bad constellation-diagram and high modulation-error-rate in DAB mode I

[Michael Kröger]

Hey everybody,

im using the MMB-tools for generating a DAB-mux in mode I and a USRP N200 for the HF-part.

I have a R&S ETL TV analyzer which is able to decode DAB and show the relevant signal parameters.

There  i have seen that in mode I the constellation diagram is deformed and the modulation-error-rate is very high.

When using the same parameters and just changing the DAB-mode to mode II the constellation-diagram looks as it should and the MER is low.

I have attached the screenshots that document this facts and I ask everyone who has a hint for solving this problem.

(Perhaps there is a bug in the source-code of the modulator which causes this error???)

Thank you everybody

Michael

[Michael Kröger]

The solution is:

You have to calibrate the URSP for the center frequency.

uhd_cal_tx_dc_offset --freq_start "center frequency" --freq_stop "center frequency"

uhd_cal_tx_iq_balance --freq_start "center frequency" --freq_stop "center frequency"

[Matthias P. Braendli]

Hi Michael,

I'm glad you found the source of this issue, and thanks for sharing your
results.

The constellation diagram looked very strange, with these rings. Did
they disappear after calibration ?

Cheers,
mpb

> --
> You received this message because you are subscribed to the Google
> Groups "mmbtools" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to crc-mmbtools...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.

[Michael Kröger]

Hi Matthias,

now the constellation diagram looks good.

I have attached the prints of the ETL.

Michael

[Mark Rosenbaum]

Hi,

we did some other Measurements, which might also be interesting for the group:
We have an USRP B100 and a N200. The N200 performs clearly better than the B100, which it should regarding the price difference.
We are not shure what causes the B100's MER to have this shape, it seems to have a connection with the internal filtering employed in the B100,
it is not related to the working frequency, it looks the same on channel 5A and 12A.
We also tested the B100 with an external clock reference, which made no difference at all. I will post some Measurements of the final System,
as soon as we have our PA and filter here.

Mark

[Matthias P. Braendli]

Hi Mark,

interesting measurements.

Do both the B100 and the N200 have a WBX ? Of the same revision ?

We had noticed some performance difference between WBXv2 and WBXv3.

mpb

[Mark Rosenbaum]

Hi Matthias,

the N200 has a WBX rev 3.0, the B100 a rev 3.1. I did a short cross-change of the WBX-Boards, they don't seem to have an influence on the quaility of the HF output signal. What I found out is that the IQ imbvalance calibration makes things worse on the B100, not much but noticeable. It seems the IQ imbalance is that small even in the unalibrated case so that the calibration doesn't do any good.
The DC offset calibration is another issue, it is strongly recommended to reduce carrier leakage and thus the MER of carriers near the center frequency.
From this point of view, the N200 is quite superior. It would be interesting to check out the new B200 how it behaves.

Mark

[Rashid Mustapha]

Mark,

I am envious of your VERY nice test set! :-)

Best regards,

Rash.

[Matthias P. Braendli]

Another idea (I'm taking advantage of you having access to nice
measuring equipment, please forgive me :-D):

Have you seen a difference with and without FIRFilter ?


The B200 might be quite different, since its RF part is another design.

mpb

> > an email to crc-mmbtools...@googlegroups.com <javascript:>.

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

> <https://groups.google.com/groups/opt_out>.

[Mark Rosenbaum]

Besides the obvious changes in spectrum, the FIRFilter seems to have no influence on the signal
(I tested with the included larger coefficient file). The attached screenshots are for the N200 but the results are also true for the B100. It seems the
DC offset calibration is a little bit off on my test machine, as there is noticeable carrier leakage in the MER-print.
No problem doing this measurements, as we have to take a lot more of them for our project documentation, which will also be published once finished.

BTW, the telnet fix works fine, now I can access the muxer-telnet-interface.

[Mark Rosenbaum]

Another thing we noticed: The MER of the B100 is strongly dependant on the output frequency. We did some measurements at 20 Mhz (with LFTX-daughterboard)
and at 60 MHz (with WBX). Both Signals look much better than the signal on Channel 12A (223,936 MHz). It seems that the PLL of the B100 suffers from
an instability, which the N200 doesn't have

[Matthias P. Braendli]

Another idea I had: both the B100 and the N200 have configurable
clocking. Do you see a difference between when changing the
master_clock_rate ?

It would be interesting to try different multiples (even and odd) of 2.048M:

e.g.

10240000 = 2048000 * 5
18432000 = 2048000 * 9
20480000 = 2048000 * 10
16384000 = 2048000 * 8
30720000 = 2048000 * 15
32768000 = 2048000 * 16


or even fractional resampling (I don't know if it's even possible) like
13107200 = 2048000 * 32 / 5

Do you have the time to try a few values ?

cheers,
mpb

[Mark Rosenbaum]

The N200 seems to have a fixed 100 MHz Clock, but I think the constellation can't get much better than that what we already have :-)
The B100 is another story. At first, ist doesn't seem to like odd multiples, fractionals aren't possible. For the base clock, ist looks like
higher values are better, so for 32 Mhz base clock, the MER RMS is 3dB below the value for 64 MHz (the default). Base clocks in a
"more binary" manner don't give better results, the often show weird behavior like segfaults.
For me, a base clock of 64 MHz seems to fit best, the sample rate should be 6,4 MHz, which results in a good aliasing image rejection and a bearable USB load.
3,2 MHz also works well, but the aliasing is clearly visible in the spectrum.
For the N200, 5 MHz sample clock are our choice.

Mark

[Matthias P. Braendli]

Ah ! you are using resampling in dabmod ! I didn't know !

That's why I set the master clock rate to a multiple of 2.048M so that
you can disable the resampler in dabmod.

I really need a test setup to try out different combinations...

mpb