Prototype 900 bit/s FreeDV blog post

[David Rowe]

Here's the scoop:

http://www.rowetel.com/blog/?p=3700

Cheers,

David

[Jasmine Strong]

Hi David,

I had a thought about your dropout problem. Since you have a relatively long delay in any case,
you might consider solving this the same way that systems like ISDB-T solve it: by using temporal
redundancy in addition to frequency-domain redundancy. By spreading your duplicative bits along the
time axis, even wideband noise or fading that causes the entire channel to drop out for some
short period will not cause the supply of bits to the codec to disappear.

I don't think a one second delay is too long for a simplex channel like HF voice. As long as there
is a substantially faster "you're tuned and someone is keyed up" indicator, I don't think this will
be a problem- possibly consider locally generated comfort noise or fake sidetone, like GSM
voice uses...

-J.

> --
> You received this message because you are subscribed to the Google Groups "digitalvoice" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.
> To post to this group, send email to digita...@googlegroups.com.
> Visit this group at http://groups.google.com/group/digitalvoice.
> For more options, visit https://groups.google.com/d/optout.

[Gregory Maxwell]

On Mon, Nov 24, 2014 at 6:30 AM, David Rowe <da...@rowetel.com> wrote:
> Here's the scoop:
>
> http://www.rowetel.com/blog/?p=3700

Very nice results.


You might be interested in some of the interesting things people are
doing with rateless coded modulation,
http://nms.csail.mit.edu/spinal/ ... mostly applicable to data rates,
SNRs, and latencies which are outside of your target, the notion of
encoding enhancement bits always in higher order modulation, for use
when the channel happens to actually have more margin sounds like it
would fit nicely with the goal you state for auxiliary carriers.

> Given the relatively short block length, is an LDPC code the best choice?

Well, the bigger consideration might be the available decoder
computational resources. Since there really is no super fast lower
quality decode of the LDPC possible.

I think I'd commented before that a half rate golay code with tightly
packed OFDM QPSK carriers can give a worst case PAPR of 3dB (there are
quite a few papers on this now if you goggle; though most of the
recent papers are about extending it to higher order modulations) ...
so there is a least one approach where the FEC can be structured to
directly reduce the original signal PAPR, effectively getting most of
the power overhead the code adds back. So there may be some to gain
in exploring how the FEC interacts with PAPR.

[David Rowe]

Hello Jasmine,

Yes I agree some interleaving in time might help. The LPDC code is
systematic (the uncoded data is included in the codeword) so this could
be sent with no interleaving and no delay for low-latency high SNR
decoding. The parity bits could then be interleaved. When FEC is "on"
you get long delay but that's better than no copy.

I too suspect that a 1 second delay on HF PTT (in particular on bad
channels) will be tolerable. There are ways it can be managed, e.g. we
can speed up or slow down the output samples, remove spaces in speech.

I think the key issue that put people off long delay before were long
sync times, and poor re-sync, and loss of sync. For example we can't
lose 3 seconds of speech if the demod loses sync. If this can be managed
then no one in a PTT conversation will be able to tell what the delay is
except at the start or end of the over.

Cheers

David

[Tony Langdon]

On 24/11/2014 7:02 PM, David Rowe wrote:
> Hello Jasmine,
>
> Yes I agree some interleaving in time might help. The LPDC code is
> systematic (the uncoded data is included in the codeword) so this
> could be sent with no interleaving and no delay for low-latency high
> SNR decoding. The parity bits could then be interleaved. When FEC is
> "on" you get long delay but that's better than no copy.
>
> I too suspect that a 1 second delay on HF PTT (in particular on bad
> channels) will be tolerable. There are ways it can be managed, e.g.
> we can speed up or slow down the output samples, remove spaces in speech.
>
> I think the key issue that put people off long delay before were long
> sync times, and poor re-sync, and loss of sync. For example we can't
> lose 3 seconds of speech if the demod loses sync. If this can be
> managed then no one in a PTT conversation will be able to tell what
> the delay is except at the start or end of the over.

There are traps with long delays (anything over 0.5 seconds), and that
is it can create a sort of "hidden transmitter" effect, in that you
don't know if someone has keyed up until you find out the other party
heard nothing but a jumbled mess, if anything, or someone else clobbered
your signal! This is common on the VoIP modes, where propagation delays
of 0.5 - 3 seconds are routine. All made worse by the "trigger fingered
operators" who not only hit PTT as soon as the last bit of RF clears the
channel, but then speak before they actually get the button in! And
it's not just hams, we get the same issue on the new trunking dispatch
channel in the firies. :) Speaking of which, a "go ahead" tone like that
used in trunking systems might be a good tool ham radios too. :)

One mitigating factor is that with HF DV, you can at least listen in
analogue mode (unless running an unattended/automated link that needs a
quiet background), and hear the signal come up, before you get the
decoded speech. However, on the flipside, latency in a DV mode reduces
the ability for people to break in between transmissions, as is common
practice on amateur radio, to join a conversation or check into nets. I
deliberately change my operating practices on high latency channels, but
many don't make the necessary adjustments.

Just thinking out loud with some of the real world issues.

--
73 de Tony VK3JED/VK3IRL
http://vkradio.com

[Jasmine Strong]

On Nov 24, 2014, at 0:41, Tony Langdon <vk3...@gmail.com> wrote:

> On 24/11/2014 7:02 PM, David Rowe wrote:
>> Hello Jasmine,
>>
>> Yes I agree some interleaving in time might help. The LPDC code is systematic (the uncoded data is included in the codeword) so this could be sent with no interleaving and no delay for low-latency high SNR decoding. The parity bits could then be interleaved. When FEC is "on" you get long delay but that's better than no copy.
>>
>> I too suspect that a 1 second delay on HF PTT (in particular on bad channels) will be tolerable. There are ways it can be managed, e.g. we can speed up or slow down the output samples, remove spaces in speech.
>>
>> I think the key issue that put people off long delay before were long sync times, and poor re-sync, and loss of sync. For example we can't lose 3 seconds of speech if the demod loses sync. If this can be managed then no one in a PTT conversation will be able to tell what the delay is except at the start or end of the over.
> There are traps with long delays (anything over 0.5 seconds), and that is it can create a sort of "hidden transmitter" effect, in that you don't know if someone has keyed up until you find out the other party heard nothing but a jumbled mess, if anything, or someone else clobbered your signal! This is common on the VoIP modes, where propagation delays of 0.5 - 3 seconds are routine.

This is why I suggested making local, very fast-to-respond comfort noise- as soon as a carrier is detected, but before audio can be decoded, the receiving software should make it sound like there is someone on the channel. GSM does this, inserting fake line noise and sometimes even breath sounds into a channel which is otherwise completely silent when no voice is being transmitted.

-J.

[Tony Langdon]

On 24/11/2014 7:43 PM, Jasmine Strong wrote:
> This is why I suggested making local, very fast-to-respond comfort noise- as soon as a carrier is detected, but before audio can be decoded, the receiving software should make it sound like there is someone on the channel. GSM does this, inserting fake line noise and sometimes even breath sounds into a channel which is otherwise completely silent when no voice is being transmitted.

Worth trying, it might just work, but definitely needs field testing. :)

[Kristoff]

David, all,


Op 24-11-14 om 07:30 schreef David Rowe:

> Here's the scoop:
> http://www.rowetel.com/blog/?p=3700

With the codec now running at these sub-1200 bps bitrates, do you think
that a AFSK-based modem for VHF/UHF is now an option too?

The 1500 bps modem I tried to hack together in c2afsk did not work, but
1200 bps AFSK is well very known technology.


Perhaps an interesting project for somebody interested in learn about
digital modems?



> Cheers,
> David
73
kristoff - ON1ARF

[David Rowe]

Hello Kristoff,

As per the blog post, one option I'd like to explore is FDM GMSK over
HF, i.e. multiple GMSK carriers rather than multiple QPSK carriers. My
understanding is that FDM-GMSK would not require a linear amplifier.

This would also double as a VHF mode, although this could be single
carrier. With a sub-1000 bit/s version of Codec 2 we could send this
through any VHF HT.

For HF, we could use the FM modulator present in many HF radios or
generate the GMSK signal completely in software and use SSB mode.

I imagine a GMSK modem has sig better than AFSK.

I'm not sure of the practical issues here, I haven't built a GMSK modem,
in particular a coherent GMSK demod (if reqd).

Cheers,

David

[Bruce Perens]

I'm confused. The mixing products of multiple GMSK carriers with
nonlinear amplification would be close in to the information carriers
and difficult to filter. - Bruce

[Matthew Cook]

You could do straight 1200bps GMSK on 10m FM I guess.

73

Matthew

VK5ZM

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice+unsubscribe@googlegroups.com.

[Matthew Cook]

Kristof,

Don't happen to have an old baycom modem (TCM3105) in your back pocket?

With codec2 running at 900-960bps and bit of HDLC bit stuffing you could feed a BELL 202 modulator directly.

Would be interesting for a quick test off the SM1000 hardware.

73

Matthew

VK5ZM

[siegfried jackstien]

1k2 Packet could be easy done with a soundcard modem software (also included
in multipsk for aprs as an example)

So no need for a (now hard to find) tcm 3105

Dg9bfc

Sigi





> -----Ursprüngliche Nachricht-----
> Von: digita...@googlegroups.com [mailto:digita...@googlegroups.com]
> Im Auftrag von Matthew Cook
> Gesendet: Dienstag, 25. November 2014 05:15
> An: digita...@googlegroups.com
> Betreff: Re: [digitalvoice] Prototype 900 bit/s FreeDV blog post

> send an email to digitalvoice...@googlegroups.com
> <mailto:digitalvoice%2Bunsu...@googlegroups.com> .

> To post to this group, send email to digita...@googlegroups.com.
> Visit this group at http://groups.google.com/group/digitalvoice

> <http://groups.google.com/group/digitalvoice> .

> For more options, visit https://groups.google.com/d/optout

> <https://groups.google.com/d/optout> .

>
>
>
> --
> You received this message because you are subscribed to the Google Groups
> "digitalvoice" group.
> To unsubscribe from this group and stop receiving emails from it, send an

> email to digitalvoice...@googlegroups.com.

[Kristoff Bonne]

Matthew,


True.

Well, actually, I am playing with some code to do AFSK modulation and demodulation on an arduino. It turns out that a atmega32u4 (the MPU found on e.g. the arduino micro) can do 9600 samples/second ADC and 1200 bps AFSK demodulation using off-the-shelf two-frequency signal-detection algorithm in about 40 % of its CPU-power.

(I do this as I want to learn more about interfacing MPUs and the analog world for our local electronics / homebrew group; using APRS as an example application.



73
kristoff - ON1ARF

On 25-11-14 06:15, Matthew Cook wrote:

Kristof,

Don't happen to have an old baycom modem (TCM3105) in your back pocket?

With codec2 running at 900-960bps and bit of HDLC bit stuffing you could feed a BELL 202 modulator directly.

Would be interesting for a quick test off the SM1000 hardware.

73

Matthew

VK5ZM

On 25 November 2014 at 06:48, Kristoff <kris...@skypro.be> wrote:

David, all,


Op 24-11-14 om 07:30 schreef David Rowe:

Here's the scoop:
  http://www.rowetel.com/blog/?p=3700

With the codec now running at these sub-1200 bps bitrates, do you think that a AFSK-based modem for VHF/UHF is now an option too?

The 1500 bps modem I tried to hack together in c2afsk did not work, but 1200 bps AFSK is well very known technology.


Perhaps an interesting project for somebody interested in learn about digital modems?

Cheers,
David

73
kristoff - ON1ARF

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.

To post to this group, send email to digita...@googlegroups.com.
Visit this group at http://groups.google.com/group/digitalvoice.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.

[Kristoff Bonne]

David,



I agree it would surely not be "the best" modem out there, but I do feel
it does have some advantages:

- you just plug it into any FM radio and it should work. No need for
radios with discriminator port. Any cheapo FM radio should do.
As Matthew said, why not an old baycom modem? :-)

- The low complexity would be a nice platform for somebody who is
interested in learning to code for digital voice.
The code for 1200 bps AFSK exists now and you can just "git clone" from
probably tens of projects on github. It should provide a good starting
point for somebody who is interested in starting out to code for codec2.



Any new bitrate requires some changes on layer 2: new interleaving
sceme, new FEC scemes, etc, so there are some interesting areas to
explore, based on the current codebase and knowledge of FreeDV and
c2gmsk. There has been some talk interframe (temporal) interleaving:
simply hack up the code and try it out!.

Even on layer 1, AFSK, I think there is some interesting work to do.

When I was working on c2afsk @ 1800 bps, I noticed that the signal-level
amplitute at 2200 Hz is a lot less then for the 1200 Hz tone. Most
implementations of AFSK ignored this. In c2afks, tried it with a fixed
correction-value.

For short packet-burst (packet-radio or APRS) that is probably the only
option there is, but for longer streams, I think it would be possible to
write some code to determine the correction-value based on the actual
characterists of the radios.


In the end, once layer 2 is good, it should run on AFSK or
single-carrier GMSK or any other layer-1 technology.. One technology
does not exclude the other. For "just a quick test" and
learning-purposes, I think AFSK should do.




73
kristoff - ON1ARF

[Matthew Cook]

Hi Kristoff,

I'l watch with interest how you go here.

I've not yet found an AFSK DSP implementation that works satisfactorily in the presence of noise.  The DSP demodulators seem to fall off the digital cliff much earlier than the older hardware based designs.

I've a small pile of fairly recent (and popular) APRS trackers that I've tested which feature various DSP demodulators and implementations.  I've not found any of them that can beat a MFJ1270b (XR2211) demodulator in the presence of real world noise.  I also bench mark them against TCM3105 and AMD7910 demodulators for good measure with similar results.  Sound card modems are no different, I've even found one or two of these start guessing which bits were corrupted in an effort to overcome some limitations.. yikes !!!   For some reason the BER performance of these old hardware chips to me seems superior, well that is what I've observed anyway, YMMV.

The biggest hurdle with AFSK is feeding your signals into and out of the radio.  You've got to watch FM radios since they also contain pre-emphasis and de-emphasis ccts, this can skew your tone amplitudes.   I don't intend to re-ignite the old packet wars on the merits of flat audio vs emphasised audio.  Just be mindful of it is all.

You should easily be able to take the AFSK code and put it back into the SM1000 hardware when ready, would certainly make a complimentary companion project to what David is working towards.

73

Matthew

VK5ZM

[David Rowe]

Hi Matt,

> I've not yet found an AFSK DSP implementation that works satisfactorily
> in the presence of noise. The DSP demodulators seem to fall off the
> digital cliff much earlier than the older hardware based designs.

Now they're just fight'n words to me ;)

Can you pls email me some low SNR AFSK samples? Interested to see what
I can do with a AFSK demod.

BTW AFSK over FM ... isn't really FSK anymore. Not quite sure what it
is from a modulation theory point of view, and what its SNR versus BER
performance should be for an ideal modem. Worth exploring though....

Another gentleman has emailed off-line about him and some friends
happily using FreeDV on UHF ... over FM radios, down to low SNRs. So
that's AQPSK, over FM ???!! Also worth looking into I think, we might
learn something.

Cheers,

David

[jdow]

David, many many years ago I ran into this phenomenon and characterized it in
terms of an ASW receiver on 137ish MHz. The USN specified an unreal low noise
figure. Then it specified the demodulator performance for the FM signal. It was
better than theory predicted. I had to do some serious reading to discover why.
It might pay to draw yourself some pictures of waveforms through an ideal
demodulator. Then imagine what that demodulator would produce if you
incorporated into it a rate limiter - something that prohibits the output from
changing level faster than X V/uS. Then try to do THAT with the DSP. (That
exercise was back in 1969 or so. It has mostly evaporated from my mind. I think
it was ratio detectors that had this trait and allowed the apparent better than
theory performance because of the "shape" of the noise down below the threshold
level. It held down to 0 dB SNR.)

{^_^} Joanne "Old age, an open mind, and experience will......"/W6MKU

[Steve]

My thoughts on the matter was to forgo the hdlc and bit stuffing. Also, send the data LSB to MSB instead of backwards, then still using NRZI for clocking, run it through a bit scrambler. Thus making it more like higher speed digital modems.

The decoder then becomes merely a symbol period counter between level transitions. Using two demodulators, one with deemphasis, one without, and you can select which one to use by using a crc in a frame header. A frame being several codec packets long.

[Matthew Cook]

Hi David,

I've always put this phenomenon down to the implementation...  :-)

I've got the gear necessary to make the recordings, thankfully my test gear for VHF/UHF makes that relatively straight forward.

The best real world test is to record the local APRS channel, that contains a unique mix of unfriendly automatic operation, collisions, weak signals, over deviated signals, noisy and fluttery signals.  If a demodulator can cope with one of these channels and hold it's own against my ancient MFJ1270b "holy grail" I'll be impressed (*grin*).

73

/M.

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice+unsubscribe@googlegroups.com.

[Matthew Cook]

Hi Joanne,

I'll show my ignorance.. what's an ASW receiver?  Anti Submarine Warfare is all I get back from Google :)

73

Matthew

VK5ZM

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice+unsubscribe@googlegroups.com.

[Steve]

here's a link to the 4X6IZ afsk modem study: A High-Performance Sound-Card AX.25 Modem

Basic low-level algorithm can probably be converted to C easily.

[Bill Vodall]

>
> Can you pls email me some low SNR AFSK samples? Interested to see what I
> can do with a AFSK demod.

A somewhat standard, for the packet TNC world, test setup is here:

http://wa8lmf.net/TNCtest/

Some additional experimentation is documented in the Dire Wolf User Guide:

http://home.comcast.net/~wb2osz/Version%201.1/User-Guide.pdf

Bill

[Matthew Cook]

Thanks Steve,

That is one software modem that I haven't seen, have put a note on the whiteboard "TODO" list in front of me.  Perhaps this implementation *might* allow me to finally retire my MFJ :)

Ah yes Direwolf, that is an interesting piece of software John has definitely done a good job.  This is the only package I've seen that will go back and look at corrupted packets and uses Error Correction techniques to guess which bits flipped (section 10.5 pg 73).  It's funny that the way he has done his TNC comparison is very similar to those that I've used.

73

Matthew

VK5ZM

On 27 November 2014 at 04:16, Steve <coupay...@gmail.com> wrote:

here's a link to the 4X6IZ afsk modem study: A High-Performance Sound-Card AX.25 Modem

Basic low-level algorithm can probably be converted to C easily.

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.

[Bill Vodall]

> Ah yes Direwolf, that is an interesting piece of software John has
> definitely done a good job. This is the only package I've seen that will go
> back and look at corrupted packets and uses Error Correction techniques to
> guess which bits flipped (section 10.5 pg 73).

One of the nice things about having multiple similar projects is that
the developers are challenged to keep up with each other. UZ7HO has
also added the bit flipping error correction. From his history file:
http://uz7.ho.ua/modem_beta/history.txt

-----
SCM v0.51b changes:
- Limited the number of bits for bit recovery operation (only for undefined
state at three different levels of threshold comparator).
WARNING! Test showed this option may cause additional erroneous CRC
collisions,
so use it at your own risk. By default is OFF

SCM v0.50b changes:
- Added single bit recovery feature
----


It looks like there's also some sort of recovery looking at historical frames.


----
SCM v0.52b changes:
- Improved MEM-recovery feature. That tries to recover corrupted frame from
the last frames buffer. It may help if the frames are often repeated, e.g.
supervisory frames. You can change the number of last packets in the INI file
(by default is "MEMRecovery=100"). Higher value may help on a very noisy path
----

73,
Bill

[Matthew Cook]

Hi Bill,

I agree.  It's good to see people back developing sound card modems again.

It's been along time since Thomas Sailer wrote his original Linux implementation, that project went quite for many years.

73

Matthew

VK5ZM

[Steve]

I visited his github at:

https://github.com/sivantoledo/javAX25

and realized that it doesn't have an executable in there. He expects you to compile it, which might not be on your short list. Anyway, I posted my compiled version of a receive only GUI part at:

https://github.com/k5okc/PacketListener

The executable is in the dist directory. My default sound card is "default [default]" on linux, so that is what is compiled in. You can change it and the sample rate with:

java -jar -Dsound.input="default [default]" -Dsample.rate="44100" PacketListener.jar

Have fun, keep that MFJ warm!  :-)

[jdow]

Got it in one - Lockheed S3A to be specific.

{^_^} Joanne

On 2014/11/25 19:20, Matthew Cook wrote:
> Hi Joanne,
>
> I'll show my ignorance.. what's an ASW receiver? Anti Submarine Warfare is all
> I get back from Google :)
>
> 73
>
> Matthew
> VK5ZM
>
> On 26 November 2014 at 12:49, jdow <jd...@earthlink.net

> email to digitalvoice+unsubscribe@__googlegroups.com
> <mailto:digitalvoice%2Bunsu...@googlegroups.com>.

> To post to this group, send email to digita...@googlegroups.com

> <mailto:digita...@googlegroups.com>.
> Visit this group at http://groups.google.com/__group/digitalvoice
> <http://groups.google.com/group/digitalvoice>.
> For more options, visit https://groups.google.com/d/__optout
> <https://groups.google.com/d/optout>.

>
> --
> You received this message because you are subscribed to the Google Groups
> "digitalvoice" group.
> To unsubscribe from this group and stop receiving emails from it, send an email

> to digitalvoice...@googlegroups.com
> <mailto:digitalvoice...@googlegroups.com>.

> To post to this group, send email to digita...@googlegroups.com

> <mailto:digita...@googlegroups.com>.

[Marciniak, Ed]

‎Limiting slew rate (time domain) is equivalent to a low pass filter (frequency domain).

The question is what we're the filters shape and phase characteristics.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.
To post to this group, send email to digita...@googlegroups.com.

[jdow]

No, it isn't. There are two different mechanisms present. Rate limiting means
you have a current limit involved in a low pass filter situation. With an ideal
LPF you can get an arbitrary rate of change on it's output simply by increasing
its input. With a rate limiter no matter how large the input signal the rate of
change on the output limits at the selected value.

Now, if you look at how this affects the signal energy you'll discover that it
removes some of the energy from sharp impulses. This is a OLD technique that
proved quite effective back in the 50s and 60s. Dig out ARRL Handbooks from that
era and you will find it discussed. It's a significant improvement over the
original simple limiter for handling noise. For me it led to an improved ability
to dig out the weak ones.

The noise on a weak FM signal is impulsive in nature. If you study the phase
domain you get a signal that drops in level and slides past the "origin"
rapidly. Since the demodulator responds to phase changes you get an impulse with
high energy in the discriminator output. If you can limit that energy, as with a
rate limiting effect, you can slice off some of the noise energy and get a
small, 1 or 2 dB, improvement in output SNR. It took me a lot of painful math to
prove this to myself. It led to the specification for the demodulator in the IF
demodulator module. (Sadly, at the moment I cannot remember what the ideal form
was. The "ratio detector" is what keeps popping into my mind. But I could not
swear it was what was selected.)

{^_^} Joanne/W6MKU

[David Rowe]

I've started taking a look at modems for VHF FreeDV, starting with 1200
bit/s FSK over FM:

http://www.rowetel.com/blog/?p=3799

- David

[Gregory Maxwell]

> Take a look at GMSK.

Make sure you're looking at a sufficiently complex GMSK decoder, when
I looked at this a few years ago the out of the box GMSK decoders
(e.g. in gnuradio) were not coherent, didn't uses viterbi to jointly
decode symbols ... etc. and seemed to leave a lot of performance on
the floor.

> Cop the performance hit and use BEL202 FSK.

Maybe, the almost complete absence of inexpensive raidos with SSB for
VHF/UHF is a barrier but on the other hand devices like bladerf and
hackrf have made direct SDR a lot more reasonable... and 7dB covers a
_lot_ of power budget difference. A dual-band 50w mobile FM radio
costs about as much as a blade rf and a 1w amplifier for it... and
that 1w is likely a lot more linear, the bladerf VCTCXO is probably a
lot more stable, so I wouldn't even be shocked if the lower power
better processing gain pair actually won out in performance.

More importantly, that 7dB loss puts the digital mode at a strong
starting disadvantage over regular FM... if the initial overhead is
enough to lose most of ability to cut-through interference and
tolerate weak signals why use digital? There are extra benefits, e.g.
being able to carry metadata,.. but by themselves I don't know if
they're that interesting.

Also running in existing NFM probably greatly limits the ability to be
spectrally efficient, even if you're able to hack a waveform that ends
up being spectrally narrow, on the RX side you're still going to have
your wide NFM demodulator lose its mind if there is a nearby signal
(it'll just lock to the wrong one, most likely), so you won't be able
to pack in more users in the same spectrum even if your signal is
narrower... and I think thats a major selling point for digital in
VHF (in UHF things like metadata and multipath handling are more
important, and spectral efficiency less so).

Plus, hardware that has direct access to the spectrum is in a position
to support experimentation with techniques like CDMA (interesting
again because more concurrency is useful in VHF); and coherent
multiple antennas.

Multiple inputs is VERY interesting for VHF/UHF due to multipath.
Since amplifier considerations already demand a constant amplitude
modulation scheme, a receiver can use the prior knowledge that the
original signal was constant envelope to blindly recover the delay
lines to coherently use multiple antennas... this scheme doesn't
require the phase between the antennas be known, so it's sufficient to
make the receivers run off a common clock (and bladerf and USRP b2xx
have refclk inputs). Simple squelch voting scheme FM receivers exist
and are widely used but I don't know if I've seen any commercial
phase-coherent multiple input receivers targeting the ham. I think
you could make a repeater with a large amount of multiple-input
processing gain for relatively modest cost (as far as repeaters go),
e.g. with 4 spaced dual polarization antennas feeding four USRP B210s
and a beefy i7 for the DSP. Maybe building something like that
commercially isn't interesting; but since it's COTS parts, it would be
a fun project for an amateur radio club. ... but it's not at all as
interesting if you're not talking a modulation scheme that can
'scales' by adding a lot of receiver intelligence.

[Steve]

Nice work on the analysis! Very well documented.

There's another thing you can take away from FSK over FM's low performance. Suppose you didn't care about communicating 100 miles, but instead maybe 15 miles.

I still remember getting my first FM stereo receiver in the mid 60's. We had just a couple stereo stations, but they were preferred because they were commercial free, and played whole albums because the DJ was busy selling soap on AM. They made all their money on clear channel AM (top-40 crap).

All I had to do was go out to the farms 10 miles away and the stereo light dropped-out, and all that was left was a crappy FM mono, but even still - the mono was better than AM in quality. Go a few miles more, and you might as well be in Kansas...

When I got my first 2m all-mode and put up a cheap antenna on the chimney, it could easily do 100 miles, but the damned thing cost me a weeks pay!  Meanwhile you could go down to the sheriffs garage and they would throw as many FM rigs in your trunk you were willing to take for $10 each.  Most became 10 mile repeaters...

So the question becomes: is there a use for short range digital voice of high quality?  Say 2400-9600 bps vocoders using discarded FM rigs. The only advantage of digital versus analog, that I can think of, is you can multiplex data and plug the thing into an IP network.

[David Rowe]

Hi Gregory,

Thank you for your interesting comments.

Re:

> More importantly, that 7dB loss puts the digital mode at a strong
> starting disadvantage over regular FM...

It's just a 7dB loss of FSK over FM versus FSK. Looking at the plots, I
think the FSK over FM scheme works OK at C/No values where FM voice
works OK (say C/No = 50dBHz). So if your radio works for FM, it will
work for FSK over FM, and a digital mode using FSK over FM will also
work OK.

Cheers,

David

[Steve]

When I got started in FM I never heard of Carson's rule. I think it was relegated to engineering level material, while we (mere technicians) used the Bessel chart.

It's a simple process actually: find the modulation index, count the number of sidebands on the chart (times 2), multiply by the modulation frequency to find bandwidth.

For example to find Modulation Index: 3000 Hz / 2200 Hz = 1.36 (8 sidebands), 3000 Hz / 1200 Hz = 2.5 (12 sidebands)

So: 8 * 2200 = 17.6 kHz BW, and 12 * 1200 = 14.4 kHz BW

Carson on the other hand suggests: 2 * (3.0 + 2.2) = 10.4 kHz BW

I don't know, those numbers always seemed unrealistic.

[David Rowe]

Hello Steve,

I used 2(5kHz deviation + 3kHz audio) = 16 kHz.

It's possible to configure the analog FM simulation with and without a
filter before the FM demod that limits the bandwidth of the "RF" signal.

I found that for high CNRs I would get harmonic distortion if I set the
filter at the input of the FM demod to less than 16 kHz. This indicates
some frequency components were being cut out by the filter, necessary
for hi-fi FM at least.

It was an interesting demo of Carson's rule.

Cheers,

David

> <https://lh4.googleusercontent.com/-0vLanqVkUfs/VI0mhtHA9oI/AAAAAAAAIY8/7-xOjPLguK4/s1600/bessel.png>

[Steve]

Maybe a lot of the difference in formula may be at what amplitude you use for the baseline. Voice isn't very critical (can tolerate more distortion). if you put the far sidebands at only 10% of the carrier voltage (-20 dB) during the deviation adjustment. I think during Carson's day, voice was the only thing considered in FM.

While AFSK is just tones, the purity of the tone for data might demand a more conservative adjustment. Maybe down -40 dB (.01 or 1%) at the desired bandwidth. I always considered on my little chart that if the sideband amplitude line was touching the baseline, that was .01. If you go up to .1 on the chart, you can usually knock out a complete sideband pair from the count. Thus for a modulation index of 1.67, 10% would be 6 sidebands times 3 kHz, or 18 kHz BW, while 1% would be 24 kHz BW.

Sorry, just morning coffee talk, nothing as technical as what you are doing!

[Marciniak, Ed]

‎Assuming you could run some test tones through a given radio in transmit and receive to find the proper alignment where certain the FM (or PM with pre-emphasis) sidebands disappear, why not simply use a GMSK modem and distort just the way required by the audio interface?

If the packet format isn't normal packet radio, there isn't much point in making it AFSK 1200bps compatible.

From: Steve
Sent: Sunday, December 14, 2014 9:20 AM
To: digita...@googlegroups.com
Reply To: digita...@googlegroups.com

[Steve]

I was playing around this morning, and the 1 kHz test signal was too confusing visually for me.

So I changed it to 3 kHz, thus +/- 5 kHz deviation of a 3 kHz test signal. This is a modulation index of 1.67.

Using my Bessel chart, that shows we should get +/- 4 sidebands for Bessel's Rule, and +/- 2 sidebands for Carson's Rule (.01 versus .1 base). Plotting this out we get this chart below. Correct me if I'm wrong, but the Y Axis is dB, so coming down -40 dB is the 60 value. -20 dB (Carson's Rule) is the 80 value.

The -20 dB cuts-off about +/- 3 sidebands. The -40 dB cuts-off +/- 1 sideband. Leaving us with +/- 4 sidebands for Bessel's Rule, and +/- 2 sidebands per Carson's Rule. All seems as predicted above.

Although, as I see it, the 3rd sideband is just barely below, or even *at* the -20 dB amplitude. So I see Carson as a very conservative rule.

The baseband graph, on the bottom, does indeed filter out everything except the +/- 2 filters. But I see the two other sidebands (each side) as being thrown over the cliff. They are very distinctive and well out of the noise.

Have fun...

[David Rowe]

Hi Steve,

Nice to see some one running my simulations!

Try a run with a CNR = 100dB and note the spurious harmonic appearing
due to the filter before the FM demod with the Carson's rule cut off.
This limits SINAD to about 50dB. Comment out that filter and the
harmonic goes away. - Hi Fi!

Carson's rule is an approximation. It's useful for our low C/N regime -
where we care more about limiting the noise bandwidth than the fidelity
lost by chopping off low energy sidebands.

Playing with GMSK now, it has similar approximations, e.g. setting the
bandwidth to where 99% of the power is.

Cheers,

David

> <https://lh3.googleusercontent.com/-rFPRUjC16gA/VJBtCzBrdbI/AAAAAAAAIZY/yaV-iLRv8J4/s1600/5-3.jpg>

[Steve]

I think an AFSK voice mode would be kind of fun. I started on a Java version, but then had to make a bunch of trips to the Great White North, and poof, I have no clue where I left off now, ha.

Ed mentioned it doesn't have to be compatible with any 1200 packet in the wild now, and that was my thinking.  I never did like all that HDLC bit-stuffing poop.

I created a little software algorithm to do the scrambling, and it sounds like the 9600 hiss.  So, bits to NRZ to scrambler, and some kind of superframe, each with multiple codec frames.  That should get the tubes (valves) nice and hot.

Obviously GMSK would be superior at the higher bit-rates, and also take less bandwidth, but there's a lot of old Analog rigs laying around in peoples garages.

Maybe two new modes for Smart Mic:  1) FDM, 2) AFSK, 3) GMSK :-)

[Marciniak, Ed]

With a suitable waveform setup using the actual modulation index for a given radio, a GMSK signal‎ should be indistinguishable from one generated directly.

Where GMSK over an FM radio is less than optimum is the wider filter bandwidth than necessary on RX. Even there a low pass filter could help some.

From: Steve
Sent: Tuesday, December 16, 2014 3:01 PM
To: digita...@googlegroups.com
Reply To: digita...@googlegroups.com
Subject: Re: [digitalvoice] Re: Prototype 900 bit/s FreeDV blog post

[David Rowe]

I've started simulating a GMSK modem and am getting pretty poor results
(4.5dB loss compared to ideal). It was suggested on the codec 2 list
that a lot of performance is being left on the floor with current GMSK
sound-card modems.

Generating GMSK is easy, a good demodulator not so easy.

I do think it's a good idea to have a sound card VHF modem that works
with legacy FM radios. I also like the idea of rolling our own modem,
and agree it doesn't have to be BEL202 FSK compatible. I think by
choosing the right waveform, we can get better performance than BEL202
FSK through legacy radios.

We could possibly dream up something that avoids the 300Hz HP filtering
on mic inputs, and so work on any $50 HT.

Cheers,

David

[kg4sgp]

Hello all,

As for a VHF modem over FM hand helds, is there any advantage to use
GMSK over a non filtered minimum shift keying? (straight MSK)

To my understanding adding a Gaussian filter to the MSK stream is an
attempt to improve the spectral efficiency of the mode (but also causes
inter symbol interference). If we are using this on a HT wouldn't the
audio filters / deviation filters limit the band width of the signal?

If they do, why not just put minimum FSK as fast as you can without
inter symbol interference? (I imagine as you hit the audio bandwidth
limit that the FM HT can put, there would be a similar inter symbol
interference effect that would happen as the minimum FSK is essentially
filtered).

Actual GMSK over RF makes sense to me, but GMSK over FM seems like were
making it harder than we need to be with out the gains of the reason why
people use the "G" in GMSK. I might be overlooking something simple
though; that happens pretty often, hi hi.

KG4SGP - Jim

[David Rowe]

Hi Jim,

In terms of BER MSK actually works slightly better than GMSK (no ISI),
but has a wider over the air bandwidth. However if we are running at
1200 bit/s (ish) that's no big deal.

Neither will pass through a mic/speaker port of a HT directly, as they
have a DC-300Hz component in the baseband waveform. However it might be
possible to tweak them to handle this, not sure.

Cheers,

David

[Matthew Cook]

HI David,

Back in the day when packet was huge and Z-calls weren't allowed on HF, we used to keep ourselves amused shunting data around.  So I thought you might like some ideas on what types of modulation have been successfully implemented over VHF/UHF packet networks.  I've just limited this to what we had running here in VK5.

• 1200.baud AFSK (Bell 202) - Unmodified Radios (either mic audio or flat)
  
• 2400 baud AFSK (Bell 202) - Radios with Flat Audio only
  
• 4800 baud PSK (HAPN) - Either True FM or tweaked phase mod radios
  
• 9600 baud PSK (K9NG/G3RUH raised cosine) - True FM radios only
  
• 19k2 baud PSK (G3RUH raised cosine) - tweaked True FM radios only
  
• 1Mbit FSK - (RAW) 10GHz 20MHz wide + modified Arcnet card
  

Here's a few URL's that you might find of interest;

http://www.amsat.org/amsat/articles/g3ruh/108.html   <<- very interesting talks about bit filtering and spectral efficiency

http://www.amsat.org/amsat/articles/g3ruh/109.html

http://www.n5dux.com/ham/files/pdf/4800%20Baud%20Modem%20Daughter%20Board.pdf

In the commercial VHF/UHF world they went the more standard FFSK/MSK/GMSK with speeds between 2400 to 9600 baud.  There are still a plethora of hardware modems around (both DSP and IC's) that can be used to implement these modems, however the hardware ones tend to be limited to the usual default baud rates (1200,2400,4800,9600) which might not suit your purpose.   As you've found the GMSK demodulators are not brilliant and need some serious guru work done to them.

There are many amateur radios on the market that sport "9600 baud interfaces" which should "technically" be capable of 9600 baud PSK (raised cosine).  So feeding these should be a doddle, the TX delay shouldn't matter that greatly provided it can be adjusted to ensure the bits and RF wobulation all meet at the right point in time.

Commerical radios (from the late 90's) are all designed to pass 9600 baud GMSK.  Getting inside these and either burying a SM1000 or wiring a SM1000 externally isn't really a problem either.

Personally I don't think you need to modulate your data stream after than 4800 baud.   If you can keep it under 4800 baud then there is still the possibility of feeding it into a 10k1 narrow band FM radio on 12.5kHz spacing....  These are becoming even easier to find/purchase for peanuts.  I'm think that the "designed for but not with principle" works in this regard....

I'd forgotten completely that the 9600bit G3RUH modems were Raised cosine PSK.  I've still got 4800 baud HAPN modems kicking around.. At least now the test gear to align them properly is not hard to find... it was much harder to find the test gear back in the early 90's.  I know where there are lots of modems stashed in cupboards should you feel like playing/familiarising yourself with them, many of us have never thrown out this gear... not after we worked out what we spent on it in the day (*grin*).

Not sure if this helps, but I was fun to walk down memory lane for at least half and hour.

73

Matthew

VK5ZM

On 17 December 2014 at 07:54, David Rowe <da...@rowetel.com> wrote:

I've started simulating a GMSK modem and am getting pretty poor results (4.5dB loss compared to ideal).  It was suggested on the codec 2 list that a lot of performance is being left on the floor with current GMSK sound-card modems.

Generating GMSK is easy, a good demodulator not so easy.

I do think it's a good idea to have a sound card VHF modem that works with legacy FM radios.  I also like the idea of rolling our own modem, and agree it doesn't have to be BEL202 FSK compatible.  I think by choosing the right waveform, we can get better performance than BEL202 FSK through legacy radios.

We could possibly dream up something that avoids the 300Hz HP filtering on mic inputs, and so work on any $50 HT.

Cheers,

David

On 17/12/14 07:43, Marciniak, Ed wrote:

With a suitable waveform setup using the actual modulation index for a given radio, a GMSK signal‎ should be indistinguishable from one generated directly.

Where GMSK over an FM radio is less than optimum is the wider filter bandwidth than necessary on RX. Even there a low pass filter could help some.


From: Steve
Sent: Tuesday, December 16, 2014 3:01 PM
To: digita...@googlegroups.com
Reply To: digita...@googlegroups.com
Subject: Re: [digitalvoice] Re: Prototype 900 bit/s FreeDV blog post


I think an AFSK voice mode would be kind of fun. I started on a Java version, but then had to make a bunch of trips to the Great White North, and poof, I have no clue where I left off now, ha.

Ed mentioned it doesn't have to be compatible with any 1200 packet in the wild now, and that was my thinking.  I never did like all that HDLC bit-stuffing poop.

I created a little software algorithm to do the scrambling, and it sounds like the 9600 hiss.  So, bits to NRZ to scrambler, and some kind of superframe, each with multiple codec frames.  That should get the tubes (valves) nice and hot.

Obviously GMSK would be superior at the higher bit-rates, and also take less bandwidth, but there's a lot of old Analog rigs laying around in peoples garages.

Maybe two new modes for Smart Mic:  1) FDM, 2) AFSK, 3) GMSK :-)

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice+unsubscribe@googlegroups.com<mailto:digitalvoice+unsubscribe@googlegroups.com>.
To post to this group, send email to digita...@googlegroups.com<mailto:digitalvoice@googlegroups.com>.

Visit this group at http://groups.google.com/group/digitalvoice.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice+unsubscribe@googlegroups.com.
To post to this group, send email to digita...@googlegroups.com.

[Ralph Brigham]

      Steve and David,

      If either of you might be interested,  I still have in my possession three rather

unique books from the ARRL: 

      Speed and more Speed Volume I;

      Speed and more Speed Volume II; and

      NOSintro  TCP/IP over Packet Radio by Ian Wade G3NRW

  

[Matthew Cook]

Ah but did you have a 2m RTTY to packet gateway that would auto deliver messages to your Siemens M100 sitting at the foot of your bed at 2am in the morning ?  

Now I just pick up my phone and read me email of a morning.,.. hmm how times have changed.

73

Matthew

VK5ZM

On 18 December 2014 at 01:49, Steve <coupay...@gmail.com> wrote:

Thanks Ralph, I've lived most of that, now I just drink beer and annoy the maid :-)

We had the AT&T Unix factory here in town, so a lot of us got 3B2 Unix boxes for cheap when they discontinued them, and ran both the 1200 baud NET (pre-NOS) and 9600 baud UUCP dialup, so we had IP to Hams while most of the world was still doing dialup UUCP.

Before Linux came out, a popular OS was Coherent. When we got some PC's we found the Coherent was pretty nice, and the 3B2's kind of became foot warmers...

Kept us off the streets anyway...

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.

[Stuart Longland]

On 27/11/14 10:43, Matthew Cook wrote:
>
> I agree. It's good to see people back developing sound card modems again.
>
> It's been along time since Thomas Sailer wrote his original Linux
> implementation, that project went quite for many years.

Indeed, so long that the original homepage for the soundmodem project is
gone.

I've mirrored the code and original page here if anyone is looking for
it: http://soundmodem.vk4msl.yi.org/

The sources came off The Wayback Machine and Debian's mirrors, so
*should* be authentic.

On the 1200-AFSK vs GMSK vs C4FM debate… I think it'll be a matter of
"code it and see what happens". The theory says all three are very
close in performance. I did wonder about using more than two states,
and I still think the PSK-based options are worth persueing too

It's good to see development of a sub-1200bps voice CODEC which may see
more robust HF communications possible and may also make possible DV
over 1200-baud AX.25.

That said, real-time DV over AX.25 I think is probably a pipe dream,
darlek-like audio isn't going to be very attractive compared to a
slightly noisy FM signal.

FM signals that would make 1200-baud AFSK struggle still make for
reasonably acceptable analogue FM contacts. Where it *could* have a use
is for a voice-mail system.

There you can afford forward erasure coding and a higher bit-rate.
Combined with the text-messaging that APRS gives you, I can see cases in
the field where such a system could be helpful for disseminating
information.

Sometimes when running a checkpoint, a message comes in and you're just
not ready to take it. Having the device be able to record the message,
play it back on receipt but then hold it for future replays in case you
missed what was said could be very useful.

Moreover, it is much easier out in the field to record a message for
transmission as voice, than to try and key the message in on a small
keypad -- and such activities are downright dangerous to attempt if in
control of a moving vehicle.

This is an aspect of DV I haven't seen tackled by the likes of
P25/D-Star/MotoTurbo/etc, and could be a handy niche for FreeDV.
--
Stuart Longland (aka Redhatter, VK4MSL)

I haven't lost my mind...
...it's backed up on a tape somewhere.

--
Stuart Longland (aka Redhatter, VK4MSL)

I haven't lost my mind...
...it's backed up on a tape somewhere.

[Stuart Longland]

On 25/11/14 17:49, Kristoff Bonne wrote:
> Well, actually, I am playing with some code to do AFSK modulation and
> demodulation on an arduino. It turns out that a atmega32u4 (the MPU
> found on e.g. the arduino micro) can do 9600 samples/second ADC and 1200
> bps AFSK demodulation using off-the-shelf two-frequency signal-detection
> algorithm in about 40 % of its CPU-power.

If you're looking for another board that has this MCU, have a look at
the Freetronics LeoStick.

It's basically a ATMega32U4 with a 16MHz crystal, an onboard tri-colour
LED and buzzer, and comes in a USB-stick form-factor.

You can use the Arduino IDE to program them, but they do come with an
unpopulated 6-pin ICSP header, which I populate then use a standard AVR
programmer to program the MCU directly.

The fact you can do 1200-baud AFSK demod with these things is very handy
to know.

Brisbane Area WICEN has an aging fleet of Kantronics KPC3s that it uses
mainly for the International Rally of Queensland, replacing some of
these with a small USB-stick sized packet TNC is rather attractive.

[David Rowe]

Last night we tested the SM1000 Beta over the air, here's the scoop:

http://www.rowetel.com/blog/?p=3846

- David

[Walter Holmes]

Major congrats there to you and Matt.

Very well done indeed.

Thanks for the update and we're all looking forward to see these guys soon.

All the best,

Walter/K5WH

[Dwight Hazen]

Would it not be better to connect the SM1000 to the data port on the Icom 706?

Mic inputs may have filters that will effect the carriers. This could be a problem on many different radios. 

Dwight WB9TLH 

On Monday, November 24, 2014 at 1:30:54 AM UTC-5, DavidRowe wrote:

Here's the scoop:

   http://www.rowetel.com/blog/?p=3700

Cheers,

David

[Gary - K7EK]

I must agree with Dwight. Recalling the early days of packet when everyone connected their TNCs to the radio mic and speaker jacks, we experienced something called a twist or skewing of tones due to filtering in those stages. Subsequently those issues went away with the introduction of data ports in latter day radios. That might be something worth considering. If I had the choice, I'd go for the data port.

Best regards,

Gary, K7EK

--

[Ralph Brigham]

Hello,

Unfortunately, I can not tell which version of the Icom 706 was being used.

If it was a MkIIG rig - yes it would have a data port; but the 706 and the MkII
did not have that item.

The audio from the radio did sound slightly tinny - just my impression.

KG4CSQ

-- ******* --
"Support Personnel are as Important as Rescuers at a Mission"
Ralph A Brigham NSS 22048RL KG4CSQ
1671 Ft Campbell Blvd
The Residences Apt #Y5
Clarksville TN 37042-3542
(931) 906-9277
--********--

[Matthew Pitts]

Guys,

Let's not forget that the goal with the SM1000 is to provide a means of using FreeDV that 1) does not require a PC and 2) is usable no matter what radio you have. Having the option of connecting it to the data input instead of the microphone and speaker connections is good, and could likely be done with a custom cable, just like any other interface.

Matthew Pitts

N8OHU

 

---

From: Ralph Brigham <ralph.ald...@gmail.com>
To: digita...@googlegroups.com
Sent: Thursday, January 29, 2015 11:30 AM

Subject: Re: [digitalvoice] Re: Prototype 900 bit/s FreeDV blog post

--********--

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice+unsub...@googlegroups.com.

[Stu Nutt]

Assuming the audio parts of the SM1000 are linear (?) I would think that this would just need some adjustment of the gain settings (is there a pot on TX and RX in the unit?) to be able to connect to the data port (for example on the 817 which is the rig it was tested on in the video, and that should be the same as my 857).

Hopefully that facility is built into the SmartMic but even if not, it should not be too difficult to add a divider circuit into the leads to adjust the gain.

Stu, G3OCR

---

This email has been checked for viruses by Avast antivirus software. www.avast.com

[Steve]

Maybe a special calibrate mode.  Where each user can request a calibrate, and the two modems will then equalize. Thus storing this measured result, and using it for all communications (until the next calibrate), or maybe have a table of calibrate values per callsign.

In this case, each would transmit a special signal, and the other modem would send back the correct (error corrected) coefficients to equalize the channel. Maybe some simple ACK/NAK protocol to get the required data across.

Another idea, might be to allow the transmit (audio) carrier to be moved down 500 Hz in Narrow Mode for legacy radios that molest the higher frequencies. Wide mode can't really be lowered in (audio) frequency much. Maybe 500 Hz is all it needs (i.e., 1000 Hz carrier instead of 1500 Hz).

[Matthew Cook]

On a standard FM rig you may find a pre-emphasis cct that could potentially skew the audio, alas on the majority of SSB rigs you thankfully don't.

We perhaps should have noted in that blog article that;

• Radio was a bog standard IC706Mk2G (generic shack in a box)
• Voice compressor was disabled
• All DSP processing, including noise blanker disabled
• Level into mic was set just *under* where the ALC would begin to reduce gain (in linear region)
  
• Radio was making 90W PEP with an average power of 17W

I know from a technical perspective that the audio path from either the microphone connector or data port is the same.  I've already explored, measured and plotted these responses previously for other reasons.

What we are seeing in the received spectrogram is the shape of the crystal filter within the IC706 (and the FT817 for that matter).  Its the combined result.  The small standard Japanese 455kHz crystal filters used in these radios are not that great, they are borderline useful even for voice.  I've got plots of the modulator in the IC706mk2G that show amplitude variations of up to 4dB in the pass band with phase variations at the edge of the xtal filter of more than 100 degrees.   This radio is definitely amateur grade, meaning it's linearity and performance is mediocre at best.     The IC706mk2G was great in its day, but its age and performance are now certainly showing.  I've got commercial/military HF radios that eat this radio for breakfast which I use for comparison.

As someone has already mentioned for FreeDV to succeed it needs to be able to cope with these "mediocre" modulators and receivers.  This does mean that the demodulator and perhaps modulator need to be able to account for the "flat-ness" of the crystal filters especially on transmit, something which David is now thinking about.

However make no mistake it worked, right out of the box and the SM1000 successfully sent and received the waveforms and a QSO was made.  So there is no doubt that the first test was a resounding success, meaning David can push on bringing the SM1000 into the world.

I for one will be watching David's blog for the first signs of pre-orders.  I'll certainly be in there from day one looking for DX contacts, these are exciting times.

73

Matthew

VK5ZM

--

You received this message because you are subscribed to the Google Groups "digitalvoice" group.
To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.
To post to this group, send email to digita...@googlegroups.com.
Visit this group at http://groups.google.com/group/digitalvoice.
For more options, visit https://groups.google.com/d/optout.

--

Matthew
VK5ZM
0487 653 245

[dc...@gmx.net]

Great news and many thanks for the nice work!

Is there or will be a list for pre orders?

Many thanks in advance

73 Attila DC0ED

[Marciniak, Ed]

I‎f the TX station generated a chirped test signal (chirped at baseband not RF) while a receiver tracked it, the receiver passband frequency and phase transfer functions would be effectively be removed. Sent back to the tx, the tx could be equalized. Then if the tx sent a chirped at rf test signal(or equalized and generated at baseband), the receiver(s) could map their own passband(s).

Assuming the tx announced it was sending a calibration and the parameters used for the chirp on the second phase, receivers could automatically map itself.
A propagation beacon could double as a phase 2 transmitter.

‎De-convolving both tx and rx transfer functions is possible with just two stations and the correct handshake.

From: Steve‎
Sent: Thursday, January 29, 2015 12:17 PM
To: digita...@googlegroups.com
Reply To: digita...@googlegroups.com

Subject: Re: [digitalvoice] Re: Prototype 900 bit/s FreeDV blog post

Maybe a special calibrate mode. Where each user can request a calibrate, and the two modems will then equalize. Thus storing this measured result, and using it for all communications (until the next calibrate), or maybe have a table of calibrate values per callsign.

In this case, each would transmit a special signal, and the other modem would send back the correct (error corrected) coefficients to equalize the channel. Maybe some simple ACK/NAK protocol to get the required data across.

Another idea, might be to allow the transmit (audio) carrier to be moved down 500 Hz in Narrow Mode for legacy radios that molest the higher frequencies. Wide mode can't really be lowered in (audio) frequency much. Maybe 500 Hz is all it needs (i.e., 1000 Hz carrier instead of 1500 Hz).

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com<mailto:digitalvoice...@googlegroups.com>.
To post to this group, send email to digita...@googlegroups.com<mailto:digita...@googlegroups.com>.

[Marciniak, Ed]

I wouldn't be surprised if you could come up with a template for common radio models that was far better than no equalization, if someone really didn't want to do a calibration once per rig.




Sent from my BlackBerry 10 smartphone on the Sprint network.
From: Steve
Sent: Friday, January 30, 2015 11:16 AM

To: digita...@googlegroups.com
Reply To: digita...@googlegroups.com
Subject: Re: [digitalvoice] Re: Prototype 900 bit/s FreeDV blog post

After sleeping on it, I think a separate mode for equalization is probably a non-starter for most hams. They just want to squeeze the trigger and not have to jump through hoops. The goal being "better than analog."

There are continuous algorithms (constant modulus algorithm) that work well for PSK. Just kind of snooping around I found this guy, who seems to have the most open-source code:

http://www.i3s.unice.fr/~zarzoso/

On Friday, January 30, 2015 at 8:29:07 AM UTC-6, NB0M Ed Marciniak wrote:
I‎f the TX station generated a chirped test signal (chirped at baseband not RF) while a receiver tracked it, the receiver passband frequency and phase transfer functions would be effectively be removed. Sent back to the tx, the tx could be equalized. Then if the tx sent a chirped at rf test signal(or equalized and generated at baseband), the receiver(s) could map their own passband(s).

Assuming the tx announced it was sending a calibration and the parameters used for the chirp on the second phase, receivers could automatically map itself.
A propagation beacon could double as a phase 2 transmitter.

‎De-convolving both tx and rx transfer functions is possible with just two stations and the correct handshake.

[David Rowe]

(subject change for clarity)

What's nice about this issue is it can probably be fixed, for example
designing a FreeDV waveform that passes through misbehaving tx filters.
It's a way to pick up a few more bits we've been leaving laying about.
Much easier than fighting absolutes like the laws of physics at -5 dB SNR!

If anyone is interested in working with me to investigate this issue,
please let me know. I can provide test samples and the DSP side if you
want to play with the radios.

- David

[jdow]

Did the chicken or the egg come first?

How do you separate two unknowns with two tests? You don't know the RX passband
characteristics. So each measurement is TX plus RX passband.

If you try three stations you can send all possible tests and maybe successfully
separate out RX and TX characteristics for all three stations. In each reception
case you have two different TX characteristics and your own RX characteristic.
For each TX you have two different RX characteristics and your own TX
characteristic. With a lot of work you can then, one hopes, solve for each in
detail. No, I don't care to work the math myself.

(This is how you measure phase noise properly when you don't have a known REALLY
REALLY good source. Test all three against each other and you can pull out the
phase noise of each if they are all about the same quality.)

{^_^} Joanne/W6MKU

[jdow]

Adaptive equalization sounds interesting to me. And you can use subtle
differences between station frequencies to cue which equalizer to use in an
N-way conversation. The side effect is that the longer Joe talks the clearer Joe
becomes.

{^_^} Joanne/W6MKU

[Marciniak, Ed]

‎What I suggested doesn't require extra turnarounds but in fact provides more than two tests.


Original Message
From: jdow
Sent: Friday, January 30, 2015 3:34 PM

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.
To post to this group, send email to digita...@googlegroups.com.

[jdow]

Explain how you split the RX effects from the TX effects. 47 measurements
between the same two stations still only measure the entire path through TX and
through RX as a single composite with no way to split them.

{^_^}

[Marciniak, Ed]

So, if you slew your receive frequency at the same time as the transmitter slewing baseband instead of rf, the receiver sees a constant frequency tone...removing the rx passband shape. That let's you extract the tx passband. when the transmitter slews rf with fixed baseband‎, you can extract the rx passband. Alternatively, you can have a fixed frequency tx, and tune the receiver, to extract the rx passband. Either way you get both. If the transmitter encoded (like frequency sweep rate/period) it's intentions or behaved in a standard way, a receiver could do what it needs to automatically provided the software could tune the radio.

In fact you could use a fixed frequency beacon and slew the receiver without an intentional transmitter. Even rfi external to the radio would work.




Original Message
From: jdow
Sent: Friday, January 30, 2015 6:07 PM

[Marciniak, Ed]

If everyone equalizes their transmitter and receiver, they'll all be "standard" signals and no changes are needed for n-way.


Original Message
From: jdow
Sent: Friday, January 30, 2015 3:42 PM

To: digita...@googlegroups.com
Reply To: digita...@googlegroups.com

Subject: Re: [digitalvoice] Tx filter and FreeDV

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.
To post to this group, send email to digita...@googlegroups.com.

[jdow]

Um, how do you slew the RX? Remote control is not particularly effective on some
models. It's very slow on others. You have the control transmit time followed by
the synthesizer lock time.

For example, on my ProII this is "pathetic" to say the least. Control is wordy
(well BYTEy) and at about 192 characters per second. AND you have to wait for
the handshake to come back before issuing the next command. AND you also have to
have the little adapter know both the appropriate command set and how to
diagnose without help what it is trying to command. That's a serious order.

Of course you will need to factor in the ionosphere effects or else have a
suitable calibration site very close to you. (NVIS propagation is notoriously
variable. Line of sight is better. BUT, sometimes even line of sight becomes
unstable on 75 meters. So your proposal is perhaps theoretically possible but
not particularly practical, methinks. It might be worth a try of some sort, I
suppose. But working across the full field of synthesized remote controllable
rigs out there is perhaps pushing things a little.

(I'm envisioning the operator speaking into the microphone, "Attention FreeDV.
Rig connection. ICOM IC-756ProII, Interface speed. 19200. End." That would be
nice until Joe tries to coach Fred through the procedure and keeps triggering
his own unit.)

{^_^}

[Jasmine Strong]

Or you could use a cheap SDR and do it in software.

-J.

[Marciniak, Ed]

‎Tune around 12MHz and pick up usb leakage or 25Mhz (ethernet clock). Normally RFI is not your friend, but any external source not moving rapidly in frequency can be used as an RX cal source for magnitude. If you have a 10 MHz source, tune to it or a harmonic.

If you don't have automatic rig control tune 25-50-100 Hz steps. That's enough frequency separation to see the jump if the fft bin is narrower.

Actually extracting phase in passband might be easier to do with at least two carriers. If you start
with one to say at1KHz baseband‎, and the second at say 900 Hz and measure the phase differential, then move, you should be able to map phase.

Original Message
From: Jasmine Strong
Sent: Friday, January 30, 2015 11:39 PM

To: digita...@googlegroups.com
Reply To: digita...@googlegroups.com

Subject: Re: [digitalvoice] Prototype 900 bit/s FreeDV blog post

[jdow]

Of course. But, we're talking about a device designed to work with user's
existing radios.

{o.o} Joanne

[Matthew Cook]

David,

I had a bit of time to think about this while driving down to the South East and back on the weekend.

In nearly all amateur (grade) radios I've played with in the past 20 years share the same xtal filter for both TX and RX.  I've also found that manufacturers tend to be cheap and keep the audio processing circuitry to a minimum, than or they can be configured so that all audio processing is able to be switched OFF.

So if we were to produce a series of tones (close to the ones that you care about with the modem) and simply measure the peak output power into a dummy load, then we should "in theory" be able to take the amplitude variation in output power as function of the crystal filter shape (assuming that the xtal filter is the major contributor).  

If this were the case going back to my first statement then we can use this "measured" crystal shape as our "first" approximate of the receive equalisation *AND* pre-distort the TX modem tone levels to equalise these also...   Take two unknowns, one measurement and solve for both; kooky.

I'm happy to try the necessary measurements of the radios I've got at my disposal.  

Keep in mind that I can also bench mark this against a set of military Rockwell Collins filters that happily pass STANAG 4539 et al.

Your thoughts?

Matthew

VK5ZM

- David

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice+unsubscribe@googlegroups.com.
To post to this group, send email to digita...@googlegroups.com.

Visit this group at http://groups.google.com/group/digitalvoice.
For more options, visit https://groups.google.com/d/optout.

[Tony Langdon]

On 2/02/2015 3:21 PM, Matthew Cook wrote:
> So if we were to produce a series of tones (close to the ones that you
> care about with the modem) and simply measure the peak output power
> into a dummy load, then we should "in theory" be able to take the
> amplitude variation in output power as function of the crystal filter
> shape (assuming that the xtal filter is the major contributor).
>

What about phase? Phase changes across the passband can adversely
affect data transmission as well.

--
73 de Tony VK3JED/VK3IRL
http://vkradio.com

[David Rowe]

Hi Matt,

Thanks for your suggestions, happy to brain storm this over a coffee
some time.

One approach we tried a few years back was to play a slow sweep through
the radio's mic input (over the modem tone range) and note variations in
the output power.

I'd also like to simulate variations in tx power, and see how x dB of
ripple, or slope, affects us. And of course characterise on different
radios, as you suggest.

It's only the tx power variations that matter. Rx filter variations
will attenuate the modem tone and noise at the same time, keeping the
SNR or each tone the same. As Tony suggests, rapid phase variations may
also cause issues.

Other possibilities are spreading the bits over several tones (e.g.
using interleaving and/or soft decn FEC), evening out any variations.

Once we get our head around the problem, we can devise FreeDV waveforms
or algorithms to handle it.

Cheers,

David

> <mailto:da...@rowetel.com>> wrote:
>
> (subject change for clarity)
>
> What's nice about this issue is it can probably be fixed, for
> example designing a FreeDV waveform that passes through misbehaving
> tx filters. It's a way to pick up a few more bits we've been
> leaving laying about. Much easier than fighting absolutes like the
> laws of physics at -5 dB SNR!
>
> If anyone is interested in working with me to investigate this
> issue, please let me know. I can provide test samples and the DSP
> side if you want to play with the radios.
>
> - David
>
> --
> You received this message because you are subscribed to the Google
> Groups "digitalvoice" group.
> To unsubscribe from this group and stop receiving emails from it,

> send an email to digitalvoice+unsubscribe@__googlegroups.com
> <mailto:digitalvoice%2Bunsu...@googlegroups.com>.

> To post to this group, send email to digita...@googlegroups.com

> <mailto:digita...@googlegroups.com>.
> Visit this group at http://groups.google.com/__group/digitalvoice
> <http://groups.google.com/group/digitalvoice>.
> For more options, visit https://groups.google.com/d/__optout
> <https://groups.google.com/d/optout>.

>
>
>
>
> --
> Matthew
> VK5ZM
> 0487 653 245
>

> --
> You received this message because you are subscribed to the Google
> Groups "digitalvoice" group.
> To unsubscribe from this group and stop receiving emails from it, send

> an email to digitalvoice...@googlegroups.com
> <mailto:digitalvoice...@googlegroups.com>.

> To post to this group, send email to digita...@googlegroups.com

> <mailto:digita...@googlegroups.com>.

[Matthew Cook]

Tony,

That's a good question.

Provided that the xtal filter is matched and loaded properly the group delay (or phase) through the mechanical xtal filter should rise adversely as you touch each side of the filter, but should be constant (with some ripple) through the middle of the pass band.  Provided we keep our modem tones in the middle of the pass band we should not be adversely affected by group delay (or phase) distortion.

To be honest I've not measured the IC706 xtal filter but will see what can be practically measured without removal of the filter.  If you've got any practical suggestions I'm all ears.

73

Matthew

VK5ZM

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.
To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice+unsubscribe@googlegroups.com.
To post to this group, send email to digita...@googlegroups.com.
Visit this group at http://groups.google.com/group/digitalvoice.
For more options, visit https://groups.google.com/d/optout.

[David Rowe]

Also the modem tones are so all we really need is constant group delay
over each modem tone bandwidth which is about 75Hz. It's OK if tone 1
and tone 16 have completely different group dealy/phase responses,

It does matter if tone 1 and tone 2 are 6dB different in tx power - the
lower one will get knocked out by noise first.

- David

On 03/02/15 09:53, Matthew Cook wrote:
> Tony,
>
> That's a good question.
>
> Provided that the xtal filter is matched and loaded properly the group
> delay (or phase) through the mechanical xtal filter should rise
> adversely as you touch each side of the filter, but should be constant
> (with some ripple) through the middle of the pass band. Provided we
> keep our modem tones in the middle of the pass band we should not be
> adversely affected by group delay (or phase) distortion.
>
> To be honest I've not measured the IC706 xtal filter but will see what
> can be practically measured without removal of the filter. If you've
> got any practical suggestions I'm all ears.
>
> 73
>
> Matthew
> VK5ZM
>
> On 2 February 2015 at 19:31, Tony Langdon <vk3...@gmail.com

> <mailto:vk3...@gmail.com>> wrote:
>
> On 2/02/2015 3:21 PM, Matthew Cook wrote:
>
> So if we were to produce a series of tones (close to the ones
> that you care about with the modem) and simply measure the peak
> output power into a dummy load, then we should "in theory" be
> able to take the amplitude variation in output power as function
> of the crystal filter shape (assuming that the xtal filter is
> the major contributor).
>
> What about phase? Phase changes across the passband can adversely
> affect data transmission as well.
>
> --
> 73 de Tony VK3JED/VK3IRL
> http://vkradio.com
>
>
> --
> You received this message because you are subscribed to the Google
> Groups "digitalvoice" group.
> To unsubscribe from this group and stop receiving emails from it,

> send an email to digitalvoice+unsubscribe@__googlegroups.com
> <mailto:digitalvoice%2Bunsu...@googlegroups.com>.

> To post to this group, send email to digita...@googlegroups.com

> <mailto:digita...@googlegroups.com>.
> Visit this group at http://groups.google.com/__group/digitalvoice
> <http://groups.google.com/group/digitalvoice>.
> For more options, visit https://groups.google.com/d/__optout
> <https://groups.google.com/d/optout>.

>
>
>
>
> --
> Matthew
> VK5ZM
> 0487 653 245
>

> --
> You received this message because you are subscribed to the Google
> Groups "digitalvoice" group.
> To unsubscribe from this group and stop receiving emails from it, send

> an email to digitalvoice...@googlegroups.com
> <mailto:digitalvoice...@googlegroups.com>.

> To post to this group, send email to digita...@googlegroups.com

> <mailto:digita...@googlegroups.com>.

[Reuven Z Gevaryahu]

My main radio is an IC706MKIIG and I too have issues where there is a difference in power between the lower and upper part of the band in FreeDV. Some of it I attributed to (and was able to reduce) by adjusting the computer volume knob, and adjusting the software volume to compensate. But some of the effect I was never able to fully squash, and I think filter issue may explain it. I have the optional 1.8k SSB Narrow filter installed, I could compare that against the normal filter, and see if the behavior is still there- I never thought to try that.

I've long know that some of the higher-end and newer rigs have better receive filters and adjacent rejection: I got to use a K3/100 on field day last year, and I'm pretty sure it could walk circles around my ICOM in tight band conditions- and I'm sure the filters are higher quality.

--Reuven (KB3EHW)

On Monday, February 2, 2015 at 6:23:15 PM UTC-5, Matthew Cook wrote:

Tony,

That's a good question.

Provided that the xtal filter is matched and loaded properly the group delay (or phase) through the mechanical xtal filter should rise adversely as you touch each side of the filter, but should be constant (with some ripple) through the middle of the pass band.  Provided we keep our modem tones in the middle of the pass band we should not be adversely affected by group delay (or phase) distortion.

To be honest I've not measured the IC706 xtal filter but will see what can be practically measured without removal of the filter.  If you've got any practical suggestions I'm all ears.

73

Matthew

VK5ZM

On 2 February 2015 at 19:31, Tony Langdon <vk3...@gmail.com> wrote:

On 2/02/2015 3:21 PM, Matthew Cook wrote:

So if we were to produce a series of tones (close to the ones that you care about with the modem) and simply measure the peak output power into a dummy load, then we should "in theory" be able to take the amplitude variation in output power as function of the crystal filter shape (assuming that the xtal filter is the major contributor).

What about phase?  Phase changes across the passband can adversely affect data transmission as well.

--
73 de Tony VK3JED/VK3IRL
http://vkradio.com

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice...@googlegroups.com.

To post to this group, send email to digita...@googlegroups.com.
Visit this group at http://groups.google.com/group/digitalvoice.
For more options, visit https://groups.google.com/d/optout.

[jdow]

You may be able to optimize this if the FreeDV could "autotune" over a 500 Hz or
so range. (You'll probably have to switch back to the stock filter to handle
wide band modes like PCALE and perhaps Digtal SSTV.

{o.o}

> On 2 February 2015 at 19:31, Tony Langdon <vk3...@gmail.com <javascript:>>

> wrote:
>
> On 2/02/2015 3:21 PM, Matthew Cook wrote:
>
> So if we were to produce a series of tones (close to the ones that
> you care about with the modem) and simply measure the peak output
> power into a dummy load, then we should "in theory" be able to take
> the amplitude variation in output power as function of the crystal
> filter shape (assuming that the xtal filter is the major contributor).
>
> What about phase? Phase changes across the passband can adversely
> affect data transmission as well.
>
> --
> 73 de Tony VK3JED/VK3IRL
> http://vkradio.com
>
>
> --
> You received this message because you are subscribed to the Google
> Groups "digitalvoice" group.
> To unsubscribe from this group and stop receiving emails from it, send

> an email to digitalvoice...@__googlegroups.com <javascript:>.

> To post to this group, send email to digita...@googlegroups.com

> <javascript:>.
> Visit this group at http://groups.google.com/__group/digitalvoice
> <http://groups.google.com/group/digitalvoice>.
> For more options, visit https://groups.google.com/d/__optout
> <https://groups.google.com/d/optout>.

>
>
>
>
> --
> Matthew
> VK5ZM
> 0487 653 245
>

> --
> You received this message because you are subscribed to the Google Groups
> "digitalvoice" group.
> To unsubscribe from this group and stop receiving emails from it, send an email
> to digitalvoice...@googlegroups.com

> <mailto:digitalvoice...@googlegroups.com>.

> To post to this group, send email to digita...@googlegroups.com

> <mailto:digita...@googlegroups.com>.

[Matthew Cook]

Good point.  

With such a small BW per tone phase is not likely to be a problem, just amplitude.

That just means we need to put our power where we need it.

73

/M.

    <mailto:digitalvoice@googlegroups.com>.

    Visit this group at http://groups.google.com/__group/digitalvoice
    <http://groups.google.com/group/digitalvoice>.
    For more options, visit https://groups.google.com/d/__optout
    <https://groups.google.com/d/optout>.

--

You received this message because you are subscribed to the Google
Groups "digitalvoice" group.
To unsubscribe from this group and stop receiving emails from it, send

an email to digitalvoice+unsubscribe@googlegroups.com
<mailto:digitalvoice+unsub...@googlegroups.com>.

To post to this group, send email to digita...@googlegroups.com

<mailto:digitalvoice@googlegroups.com>.

Visit this group at http://groups.google.com/group/digitalvoice.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "digitalvoice" group.

To unsubscribe from this group and stop receiving emails from it, send an email to digitalvoice+unsubscribe@googlegroups.com.
To post to this group, send email to digita...@googlegroups.com.

[David Rowe]

Daniel, VA7DRM, and I have been doing some tests using GMSK, Codec 2,
and real radios. It appears we are getting at 10dB gain over Analog FM
and 1sts gen DV systems:

http://www.rowetel.com/blog/?p=3856

- David

------------------------------------------------------------------------------
Dive into the World of Parallel Programming. The Go Parallel Website,
sponsored by Intel and developed in partnership with Slashdot Media, is your
hub for all things parallel software development, from weekly thought
leadership blogs to news, videos, case studies, tutorials and more. Take a
look and join the conversation now. http://goparallel.sourceforge.net/
_______________________________________________
Freetel-codec2 mailing list
Freetel...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

[David Rowe]

[si...@mungewell.org]

> Daniel, VA7DRM, and I have been doing some tests using GMSK, Codec 2,
> and real radios. It appears we are getting at 10dB gain over Analog FM
> and 1sts gen DV systems:

“we own the stack”. Codec open, modem open, protocol open.

You should put that on a T-shirt...

Great news, and congratulations on your achievements.
Simon.