JT65 SNR (signal-to-noise ratio)

[SP2BPD]

This is not about Linrad but I do hope somebody will be able to answer my question. It's about SN ratio in JT65. 

Given useful (copyable) signals have SNR somewhere between -24dB and 0 dB, I guess JT65 has its own definition of signal-to-noise ratio:

SNR = 10 log (S/(S+N)), where S is power of signal and N is noise power in the bandwidth of 2.5kHz?

Am I right?

If positive: Could you give any link with "official" confirmation of my hypothesis? :-)

Regards

Piotr, SP2BPD

[Joe Taylor]

Hi Piotr,

WSJT modes have always specified signal-to-noise ratio using a 2500 Hz
reference noise bandwidth. Thus,

SNR = 10*log10(S/N)

... where S is signal power and N is noise power in bandwidth 2500 Hz.

For example, see here in the WSJT-X User Guide:

http://www.physics.princeton.edu/pulsar/K1JT/wsjtx-doc/wsjtx-main-1.7.0-rc3.html#_standard_exchange

"Signal reports are specified as signal-to-noise ratio (S/N) in dB,
using a standard reference noise bandwidth of 2500 Hz."

In general, noise power is measured in a bandwidth smaller than 2500 Hz,
using the "baseline" of the measured spectrum close to the signal
frequency. The value is then scaled up to what it would be in 2500 Hz
bandwidth assuming a flat spectrum.

-- 73, Joe, K1JT

[SP2BPD]

Joe,

thanks for your prompt answer.

Some time ago I performed a kind of "virtual experiment" adding white noise to real JT65 signal. The real signal was recorded in Linrad, then played back and sent to DL4YHF's Spectrum Lab.

In Spectrum Lab  noise of different levels have been added and such a mixture has been decoded by JT65-HF soft. I've noticed that for SNR less than -10dB JT-65HF reports were decreasing 3dB for every 3dB more noise added in SpectrumLab.

For signal-to-noise ratios bigger than -10dB this dependance was not linear. This led me to hypothesis from my first email. So I must think it out again.

By the way. I'm going to experiment with phased antennas in lower HF bands. I'll record JT-65HF signals on 2 channel receiver. Then I will play my recordings back again and again with different phase shift and amplitude balance ratios between the channels. I hope to verify what 4nec2 models say about Receiving Directivity Factor of properly phased antennas. Expect to see JT65 3 dB SNR improvement where modelling says RDF of phased loops is 10dB against 1 loop RDF equal to 7 dB. What do you think about my methodology?

Regards

Piotr, SP2BPD

[Joe Taylor]

Hi Piotr,

On 12/14/2016 1:29 PM, SP2BPD wrote:

> Some time ago I performed a kind of "virtual experiment" adding white noise
> to real JT65 signal. The real signal was recorded in Linrad, then played
> back and sent to DL4YHF's Spectrum Lab.
>
> In Spectrum Lab noise of different levels have been added and such a
> mixture has been decoded by JT65-HF soft. I've noticed that for SNR less
> than -10dB JT-65HF reports were decreasing 3dB for every 3dB more noise
> added in SpectrumLab.

I can't tell you much about JT65-HF. That's not my software.

> For signal-to-noise ratios bigger than -10dB this dependance was not
> linear. This led me to hypothesis from my first email. So I must think it
> out again.

Once again, you might want to look carefully at the passage to which I
directed you in the WSJT-X User Guide:

http://www.physics.princeton.edu/pulsar/K1JT/wsjtx-doc/wsjtx-main-1.7.0-rc3.html#_standard_exchange


"Signal reports are specified as signal-to-noise ratio (S/N) in dB,

using a standard reference noise bandwidth of 2500 Hz. ... JT65 reports
are constrained to lie in the range –30 to –1 dB, and values are
significantly compressed above about -10 dB. JT9 supports the extended
range –50 to +49 dB and assigns more reliable numbers to relatively
strong signals.

> By the way. I'm going to experiment with phased antennas in lower HF bands.
> I'll record JT-65HF signals on 2 channel receiver. Then I will play my
> recordings back again and again with different phase shift and amplitude
> balance ratios between the channels. I hope to verify what 4nec2 models say
> about Receiving Directivity Factor of properly phased antennas. Expect to
> see JT65 3 dB SNR improvement where modelling says RDF of phased loops is
> 10dB against 1 loop RDF equal to 7 dB. What do you think about my
> methodology?

For reasons outlines above, I recommend using WSJT-X and JT9 rather than
JT65-HF and JT65. In any case you will need to watch out for many
effects that can cause systematic deviations in estimated SNR values.

-- Joe, K1JT

[SP2BPD]

Joe (and All who can contribute),

As JT9 signals are rare on 1.8 MHz I've appointed three hams who will transmit test signals for me.

Before I start the experiment I want to be sure my methodology is correct. The RDF (receiving directivity factor) of MW/HF antennas is described by W8JI as “the ratio of response in the desired direction … to the overall gain of the antenna” or “difference between gain in desired direction and average gain”. The goal of my experiment is to measure RDF difference between arrays made up a pair of loops (arranged endfire, with the axis on direction of incoming signal). Actually I will use only one pair of loops but I will form different arrays afterwards, replaying two-channel recordings with different phase shift between the channels.

SNR will be measured for each array. Difference between SNRs is supposed to be equal to difference of RDFs.

In my understanding three conditions shall be fulfilled: 

1/ uniform “density” of noise over whole hemisphere above the antenna

2/ lack of other specific noise/signals in the received bandwidth

3/ dimensions of my loops big enough to provide level of ionospheric noise well above the noise level of preamplifiers.

I have no influence on condition 1.

To satisfy condition 2 I will have to avoid frequencies on which there is anything audible or visible (on a waterfall) except of test signal.

Am I right? Any more hints, suggestions?

Kind regards

Piotr, SP2BPD

[Joe Taylor]

Piotr --

On 12/21/2016 7:13 AM, SP2BPD wrote:
> Joe (and All who can contribute),
>
> As JT9 signals are rare on 1.8 MHz I've appointed three hams who will
> transmit test signals for me.
>
> Before I start the experiment I want to be sure my methodology is
> correct. The RDF (receiving directivity factor) of MW/HF antennas is
> described by W8JI as “the ratio of response in the desired direction …
> to the overall gain of the antenna” or “difference between gain in
> desired direction and average gain”. The goal of my experiment is to
> measure RDF difference between arrays made up a pair of loops (arranged
> endfire, with the axis on direction of incoming signal). Actually I will
> use only one pair of loops but I will form different arrays afterwards,
> replaying two-channel recordings with different phase shift between the
> channels.
>
> SNR will be measured for each array. Difference between SNRs is supposed
> to be equal to difference of RDFs.
>
> In my understanding three conditions shall be fulfilled:
>
> 1/ uniform “density” of noise over whole hemisphere above the antenna
>
> 2/ lack of other specific noise/signals in the received bandwidth

There should be no need for this requirement, so long as your passband
(of order 2-4 kHz wide) is not filled with signals. As I stated here
before:

"In general, noise power is measured in a bandwidth smaller than 2500
Hz, using the "baseline" of the measured spectrum close to the signal
frequency. The value is then scaled up to what it would be in 2500 Hz
bandwidth assuming a flat spectrum."

Just be sure that there is some good "baseline" devoid of signals.

Be sure to turn AGC off.

-- Joe, K1JT

[SP2BPD]

Joe,

>There should be no need for this requirement, so long as your passband

> (of order 2-4 kHz wide) is not filled with signals.

I'm going to perform my experiments in mid-January. Since they are other hams involved and since receiving array will be located in snowy and frosty field (I don't like cold ! ) I want to be prepared as perfectly as possible.

 

I'm afraid a condition of 2 kHz passband free of any signals might be difficult to meet. In a case I will see (hear) a specific noise or unwanted signals close to test signals: how far off that noise should my colleagues set their transmitters?

 

You’ve also written: "In general, noise power is measured in a bandwidth smaller than 2500 Hz, using the "baseline" of the measured spectrum close to the signal frequency.  The value is then scaled up to what it would be in 2500 Hz bandwidth assuming a flat spectrum."

 

My feeling is this that “spectrum close to the signal frequency” for JT9 is approximately 170 Hz wide. Am I right?

 

If positive: I’m going to use a rule of thumb for test signals to keep at least 100 Hz margins below and above the signal clear of any specific noise or unwanted signals. I want only pure ionospheric noise in +/-100 Hz vicinity of test signal. Am I right?

 

Anyway. Thank you for your patience.  :-)

 

Piotr, SP2BPD