identification of spurious tx components

[ks1h]

Got started looking at this because my Peaberry CW note produces lots of buzzing in my main station receiver. When I looked at the output on my scope I got this:

peaberry CW ouput 10uS/div

peaberryCW_txmit.jpeg (23.95 KiB) Viewed 9751 times

The modulation here is about 10uS period or, more precisely, 96kHz as determined from my station receiver. I first thought this was related to the 96kHz peaberry audio rate but this is not the case as the modulation frequency (but not the amplitude) changes as the difference between LO and tune frequency changes. I can change this now that I can change and rebuild the source code. In fact, the modulation frequency is 4x the difference between LO and tune (in the unmodified PeaberryCW this is 24kHz). I feel certain now that this is due to the way the SN74CBT3253 IQ modulator works. Each cycle of the LO clock samples the I and Q lines 4 times: I, Q, -I and -Q. The transmit signal starts with I=1 and Q=0 and then phase rotates toward I=0 and Q=1, and so on. As long as the phase is such that I or Q is near maximum, the output of the mixer will be maximum. However, when the phase is near 45deg, I will be cos(45)=0.707 and Q will be sin(45)=0.707. The output of the mixer will then be only 71% of the maximum. This happens for values of phase = 45, 135, 225, and 315deg; i.e., four times per audio cycle and leads to the modulation frequency of 4x the difference between LO and time. Here is a calculation of the mixer output for a LO of 10MHz:

Calculation of mixer output with 24kHz audio

Envelope.JPG (35.32 KiB) Viewed 9751 times

and an expansion of the low amplitude part:

Expanded mixer output

Envelope_binning.JPG (21.98 KiB) Viewed 9751 times

I thought maybe if I made the LO-tune freq very small then this modulation could be made with a longer period than typical keying dots and dashes but I could not get the Peaberry to xmit with LO-tune differences less than about 50Hz. With this you just get a monstrous 50Hz buzz on the CW tone. More work to be done on this as I get time.

73, Ken KS1H

[ks1h]

I've taken a brute force approach to try to reduce this effect: create a 4x frequency component out of phase with the one showing up in the output and use it to modulate the sine wave that PeaberryCW sends out. To do this I couldn't use the 24kHz LO-tune diff that Peaberry CW uses as the PEABERRYRATE of 96kHz doesn't give enough granularity to make a good signal at 4x the frequency. So I reduced the LO-tune diff a factor of four to 6kHz. This parameter was hard-coded in only (I hope!) four files: demod.cpp, cat.cpp, settingsform.cpp and audio.cpp. To make it easier to change, I defined it in dsp.h as LOTUNEDIFF. The transmit sine wave alterations are done in the last routine of audio_win.cpp. I found the the correction was strongly amplitude dependent so different correction factors are used for low output and for high output. I tuned the phase and amplitude of the correction to minimize the output 4x component. It worked fairly well and I can barely see the 4x component now on my scope (I don't have a spectrum analyzer, though). I can still hear some some extra tones on the CW carrier with my station receiver; not sure what this is due to. Maybe someone with a spectrum analyzer can look at the signal and see how clean it is.

P.S. will have to find out how to upload .h and .cpp files as the website refuses to do it.

73, Ken KS1H

[ks1h]

Here is a zip of the files

[W4MMP]

Hi Ken,
Would you provide your test conditions? Just the basic stuff like which model and band your testing? (I have the 30/20/1 7/15 and 80/75/60/40). I have a basic USB scope and it tops out at the 20M band. I would like to examine the output with my scope to see if I can observe what you are seeing.

This will be a nice diversion for me. I'm getting no where with my work on software. Nothing I attempt to compile on the RPI 2 works except for Fldigi so I'm putting my software efforts aside for a while.

[ks1h]

I built the 60/40/30/20 version and was testing on 40 Mtrs. I'll be very interested to find out if you see the same thing.

73, Ken KS1H

[W4MMP]

Hi Ken,

I performed some observations this evening.
Test Conditions:
HDSDR: CW Mode - Band 40M
Peaberry: 80/75/60/40
Power Out: 1W

Please find below screen shots representing various LO/Tune offsets. I started with a 1KHz offset. Then moved to a 10KHz offset and then added 10KHz increments until HDSDR would not permit me to increase the split any further. I do not understand enough about what I am seeing but it is line with what you have been observing.
No apparent ripple.

Screenshot 2015-04-06 21.23.39_cr.jpg (252.5 KiB) Viewed 9680 times

Still no apparent ripple.

Screenshot 2015-04-06 21.24.00_cr.jpg (248.42 KiB) Viewed 9680 times

Ripple now becoming noticeable

Screenshot 2015-04-06 21.24.34_cr.jpg (243.62 KiB) Viewed 9680 times

I will post more screen shot in another post. The forum only allows three attachments.

[W4MMP]

More screen shots.
Ripple very pronounced.

Screenshot 2015-04-06 21.24.55_cr.jpg (242.86 KiB) Viewed 9680 times

Maximum LO/Tune offset (48KHz)

Screenshot 2015-04-06 21.25.27_cr.jpg (245.2 KiB) Viewed 9680 times

[ks1h]

Hi Ron,
I saw two effects. One is a fixed frequency 96kHz modulation of the carrier where the modulation gets greater with increasing LO-tune difference. This is the one you are showing. The other one I saw with PeaberryCW where the modulation frequency is 4X the LO-tune diff but the modulation amplitude is approximately fixed. I think they are two different effects.

The first one I think may be a difficulty in the pcm3060 to create a harmonic-free sine wave near the Nyquist sampling limit (48kHz for a 96kHz sampling rate). The pcm3060 documentation hints at this: "This filter is not enough to attenuate the out-of-band noise to an acceptable level for many applications in general. An external low-pass filter is used if further out-of-band noise rejection in required."

The second effect I think is due to sampling in the Sn74CBT3253 IQ mixer as I've described in previous posts in this thread. With PeaberryCW, the pcm3060 makes a very nice looking sine wave going into the Sn74CBT3253 inputs (at least up to 24kHz LO-tune diff) but the output of the Sn74CBT3253 is modulated at 4x the LO-tune difference. With unmodified PeaberryCW, this modulation shows up as 96kHz, the same frequency as in the first effect. The 4x modulation becomes unambiguous (i.e., shows up at different frequency than 96kHz) when the LO-tune difference is changed to something lower than 24kHz. I haven't looked at this effect with HDSDR because, frankly, it is a black box and who knows what it is doing. At least with PeaberryCW, one knows exactly what the software is doing.

73, Ken KS1H

[ks1h]

Well, this 4x LO-tune component has been a fascinating one to look into. I finally managed to get a Spectrum Lab spectrum of my station receiver audio and then tried unsuccessfully to eliminate the 4x component by adding an out-of-phase 4x modulation to the PeaberryCW signal (even though it could be minimized on my scope). So I went back to basics and fleshed out my computer model (described above in this thread) of the SN74CBT3253 output. To my great surprise, when I ran a FFT of the output which had an obvious 4x modulation, there were NO 4x sidebands to the signal. Instead, there are strong odd harmonics which are not quite odd multiples of the fundamental. I say not-quite-odd-multiples because they are displaced from an exact multiple by 4x the LO-tune amount. These beat with the fundamental to produce the apparent 4x modulation. I checked with my station receiver and verified that these not-quite-harmonics are, indeed, there and at the right frequency. So trying to get rid of this by adding a 4x modulation will, I'm sorry to say, not work. What will work, and I've verified this by chopping off the harmonics in my FFT, is to have a strong enough low pass filter to eliminate the odd harmonics. I'm a little surprised that the built-in Peaberry filter doesn't do this. Maybe I didn't build it quite right. In any case, a good antenna tuner should do it also.

73, Ken KS1H

[W4MMP]

Hi Ken,
Excellent analysis and work!. But the question still remains of the "hum" you can hear on your signal. While a low pass filter can eliminate the offset odd harmonics, how is that going to eliminate the hum on the fundamental?
(What did I miss?)