Pages: 1 2 3 [4]   Go Down

Author Topic: Baseband FFT system - any h/w and s/w experience?  (Read 703 times)

VK6HP

  • Member
  • Posts: 1222
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #45 on: October 17, 2021, 10:56:51 PM »

Certainly some good results from the low cost crystals, Jeremy.  I am encouraged, although mine will be a while turning up.  I think I ordered 10 of each frequency with the intention of choosing the closest 5 for the notch filter.  I recalled this morning where I saw the idea of "doubling up" the shunt crystals: it was in the July/August 2008 edition of QEX, in Wes Hayward's (W7ZOI) article on oscillator evaluation with a crystal notch filter.  Looking again at that article, I see he achieved about 85 dB null depth at ~10 MHz with 5 crystals, but his crystals may have had lower ESR. 

For most of my LO applications between 10 and 70 MHz I think the SML-01 in divider mode with a 20 kHz noise floor of -148 dBc will be a good enough work horse.  But better sources are always good to have and are, of course, essential in more advanced applications.  Wenzel's application notes are well worth reading, too, and I notice they have a number of useful schematics including a low-noise post-mixer amp for baseband work.  For the 1/f and other low-frequency noise processes nothing beats some measurements, as both you and HH75 note.

I'm professionally acquainted with the team here in Perth who designed the Poseidon Scientific Instruments (now Raytheon) ultra low noise sources although I don't see myself getting hold of a sapphire resonator any time soon.  But they were/are the ultimate in terms of close-in phase noise and, being 10 or 20 dB better than the Wenzel competitors in radar applications, allowed you to know that your seconds were numbered as the Patriot (or whatever) missile approached! 

73, Peter.

Logged

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #46 on: October 18, 2021, 10:31:50 AM »

Thanks Peter. Lots of useful info as ever in Wes Hayward's article. Looking at his 10MHz filter he has the same filter topology as me but I think he has opted to match up to a higher filter impedance to make it easier to get a deep notch. The filter Z vs crystal ESR is a higher ratio so the notch can be really deep with a a higher Z filter.

In my case I found that this has the penalty of a wider bandwidth at the top of the notch. To help sway me further towards a 50R design I also have a huge bag of Toko 10k 820nH coils here and these were ideal for the 90degree sections at 9.83MHz if the filter was designed for a Z of 50R. I was therefore fairly committed to a 50R design if I wanted something quick and cheap. I would probably need about 8 crystals to match the notch depth Wes achieved with just 5 crystals. It is tempting to add another section but I'm not sure I need it in my case. I have plenty more of the Toko 820nH coils and I'm slowly using them up. With my notch I think I can measure reliably at a 1kHz offset but this looks marginal with Wes' notch filter as it is much wider at the top of the notch.
Logged
Regards
Jeremy

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #47 on: October 18, 2021, 11:52:10 AM »

I had a go at making the SSB crystal filter on a spare scrap of PCB but the results aren't as good as I'd hoped. I couldn't find 8 matched crystals amongst the remaining ones I had left. There is nothing to adjust in the design so it is a case of build and hope.

You can see there is a slight droop on the high side of the passband and this must be to do with the spread amongst the crystals. Otherwise it was very close to the simulation. It is very risky to design the filter like this. What I should have done was match the crystals first and then taken s-parameters of the one in the middle of the spread of matched crystals. Then designed it on the simulator.

I'm going to buy another 20 crystals and try again at a later date.

Logged
Regards
Jeremy

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #48 on: October 18, 2021, 02:16:00 PM »

I found that padding a couple of the large caps by +5% improved it a lot so some of the droop may be down to component tolerances. It is really close to the simulation now. This design deserves a bit more time spent on circuit optimisation to make it repeatable. 

I also found some more SRA-3 mixers on an old dev board and this meant I could build up another quadrature network although this works at 10.7MHz rather than 9.83MHz. At the moment it only works well if I include at least one Agilent ESGD vector sig gen as it has a variable phase control to accurately achieve quadrature. Both sig gen sources have to share a common 10MHz reference.

It seems to work quite well although I haven't added a low noise at at the DC port of the mixer. The last time I tried this I had issues with ground loop interference and the way to stop this seems to be to use a 1:1 isolation transformer at the sig gen outputs. Without this the IF port fills up with ground loop signals and these are mainly harmonics of 50Hz.

Without the low noise amp at the detector output port of the SRA-3 it is limited to about -145dBc/Hz. I'm hoping to see much lower with a low noise amp circuit.


Logged
Regards
Jeremy

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #49 on: October 18, 2021, 03:18:47 PM »

Wenzel's application notes are well worth reading, too, and I notice they have a number of useful schematics including a low-noise post-mixer amp for baseband work. 
73, Peter.

I did a quick google and found this circuit from Charles Wenzel. The performance looks to be very good but it does seem to be quite a complex circuit.

https://www.wenzel.com/wp-content/uploads/lowamp.pdf

I suspect KE5FX and SM5BSZ will have some decent circuits for this as well. I will probably start out with a common base BJT with a low bias current and then follow this with an emitter follower. If this doesn't work well then I'll see what proven circuits are out there.

The team in Perth would probably be the ultimate consultants for your baseband FFT project. Useful people to know!
Logged
Regards
Jeremy

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #50 on: October 20, 2021, 10:50:34 AM »

I had access to the E5052A signal source analyser today so I swapped the 9.83MHz crystal for a nice IQD 10MHz crystal in my 50R MMIC based 'Leeson Trainer' circuit. This crystal has low ESR but the loaded Q is only 6200 in the MMIC feedback circuit.

However, I ran it at a higher resonator drive power of about 2.8dBm into the crystal by setting the loop attenuation to 5dB. Leeson's equation predicted the phase noise plot below and the real measurement is also given below. This shows 7.8dBm Pout but this is after the MMIC and the equal power splitter. There is 5dB loss in the attenuator so the resonator input power is 2.8dBm. There's about 1dB loss in the crystal. The MMIC has a noise figure of about 5.5dB here and there will be some compression effects to factor in as well. I've factored all this into the Leeson plot below.

I think the phase noise is probably too low for the E5052A to measure at 1kHz offset. I used a correlation of 10 to get the plot below. This is probably the limit of the E5052A. However, the rest of the plot was in quite close agreement and I put in a flicker corner of 1500Hz. I do feel it is possible to design low phase noise crystal oscillators cheaply using a basic MMIC circuit although I am probably driving the crystal quite hard here. I did try increasing the attenuation but there is only 3.5dB gain margin in the loop at 5dB attenuation. So it stopped oscillating with 8dB attenuation in the loop.

Getting close to -170dBc/Hz at 50kHz offset is quite good I think. I'm a bit more confident about designing test oscillators like this. I think the crystal ageing will be poor at this high drive level but that doesn't really matter in this case.

It should be possible to make a suite of these oscillators at various frequencies to either use as a low noise LO or as a gold reference for that frequency. This would complement any homebrew phase noise measuring system I think.

Sadly, I had to do this at my workbench and there is a lot of local interference from nearby equipment where I work. So there are a few spurious terms visible on the plot. The spurious term at 60kHz always seems to be there and it may be an internal spurious in the E5052A. There is another E5052A I could try at a later date.


« Last Edit: October 20, 2021, 10:56:59 AM by G0HZU »
Logged
Regards
Jeremy

VK6HP

  • Member
  • Posts: 1222
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #51 on: October 21, 2021, 02:37:32 AM »

That's a good result, Jeremy.  It certainly looks as though a similar oscillator would be a decent bench standard, and would be ideal for characterizing the phase noise behaviour of e.g. the majority of common 10 MHz frequency standards.  It'd be interesting to see how repeatable the characteristic is but, on the other hand, just having something respectable that can be measured initially on a lab instrument would on its own be very useful.  It's of order 10 dB better in close-in noise relative to the Crystek and Golledge units we were discussing, although to be fair I guess we should take a few dB of those units' plots to account for a division to 10 MHz.

My crystals seem to be a casualty of the Fedex delivery network and, looking at the web predictions, I'm not expecting them until next week.  In the meantime I've been trying to keep gainfully occupied by pushing on with construction of a 136 kHz power amp.

73, Peter

Logged

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #52 on: October 21, 2021, 03:20:51 PM »

Yes, I was quite pleased to see the oscillator perform this well. In terms of low cost options for the FFT based back end I tried playing with the Analog Discovery 2 and also the RSP1A SDR.  The spectrum analyser function in the AD2 is very basic and very limited so I quickly gave up on that. However, the little RSP1A performed really well with the 9.83MHz notch filter. I used the  spectrum analyser program written by Steve Andrew.

It managed to display the noise of the Mi2019 on a 10kHz span although I think there is an issue with the FFT window function because it doesn't seem to display the noise power accurately. However, if this is ignored, it can be used for relative measurements. I was surprised it worked so well for a £100 SDR. I have also explored using it with a noise source for Y factor noise figure testing and it did extremely well for this too. Certainly good enough for casual measurements of amplifiers with noise figures above (say) 5dB.

There's so much choice out there and the prices vary a lot too. For the ultra high performance stuff up to 100kHz or so it might be worth asking on DIYaudio. There are some very experienced forum members on there. I've not registered for it but I do sometimes end up there on a google search and some of the content is very good.

Logged
Regards
Jeremy

VK6HP

  • Member
  • Posts: 1222
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #53 on: October 28, 2021, 05:45:22 PM »

The supply chain problems that everyone talks about have definitely hit my Digikey crystal delivery but I managed to get hold of five 11.06 MHz crystals from RS Components in Australia (stock 672-0208).  These looked quite good on the VNA, with fairly low ESR and frequency spreads of <800 Hz in their series resonance.  The arithmetic mean resonant frequency of the set was 11.05412 MHz. 

I put together a 5-crystal notch filter, along the lines of the Hayward design (50:200 ohm i/o transformation).  My spectrum analyser is with the agent to have its real-time module checked out so I pushed a nanoVNA, nanoVNASaver software and a Touchstone file viewer to produce some decent results, although don't get hung up on absolute transmission values below -70 dB.  The un-tweaked response is shown below:



While there is room for optimization, the raw performance is pretty good in the notch application.  Still, the minimum usable offset is a little less than 2 kHz, limiting the usefulness in some applications while being adequate for many.  One thing the measurements remind us of is the poor return loss around the notch: a short is still a short however one transforms it!  Again though, for many applications it's not a big issue.

Incidentally, the 11.06 MHz crystals look considerably better than some RS 10 MHz ones I ordered at the same time, the 11.06 MHz units having noticeably lower ESR and frequency spread.  It will be interesting to see how the Digikey ones shape up when they arrive.

73, Peter.




Logged

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #54 on: November 06, 2021, 05:40:43 PM »

The notch depth of your filter is very good and fairly wide too. I found that I could optimise the return loss close the notch by adjusting the slugs in the Toko 10k coils on my notch filter.

I had a good result last week when I dug out a 102.4MHz VCXO evaluation circuit. This is an old dev board made by a colleague at work and I rescued it from disposal a few years ago. It has a PIC controlled ADF4001 PLL circuit and the option to lock to either an on board 10MHz TCXO or an external reference oscillator. The original PIC code was not optimal so I rewrote it to give a nicer phase noise response. The result was really quite good. See below.



The phase noise is about -140dBc/Hz at 1kHz offset. This evening I tried using it as a reference clock to drive an AD9954 DDS. I then programmed the DDS to my 9.8279MHz notch frequency and the phase noise via the notch was about the same as the AD9954 datasheet at 9.5MHz. However, I tried putting 47uF at the DAC_BP pin of the DDS and this cleaned up the phase noise even more. At 9.8279MHz the DDS output appears to be delivering about -146dBc/Hz at 1kHz offset and -152dBc/Hz at 10kHz offset. This improves to about -155dBc/Hz at 100kHz offset.

This is close to what I was hoping to achieve. There are other DDS chips that could deliver even lower phase noise and this might be worth exploring.
Logged
Regards
Jeremy

VK6HP

  • Member
  • Posts: 1222
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #55 on: November 07, 2021, 01:07:19 AM »

That's a good result!  -150 dBc/Hz at 20 kHz offset, at a carrier of 105 MHz, is very competitive with good commercial systems.  It's also interesting to see the notch filter approach is proving useful in getting phase noise readings consistent with a direct measurement at nearby frequencies.

My 10 MHz series resonant crystals (CTS MP100-E, 30 ppm) arrived and, in an economy of effort, I simply wired the best five in parallel with the previous 11.06 MHz units, with no change to the LC synthetic quarter-wave line.  The mean series resonant frequency was 10.000100 MHz.

The previous 11.06 MHz response was unchanged and no spurious responses were introduced.  Again using the nanoVNA, and with no tweaking, the raw notch characteristic is shown below.



The notch depth is very good and with the nanoVNA frequency within 1 Hz or so of my 10 MHz GPSDO, it's clear that phase noise measurements with offsets greater than about 1.5 kHz would be quite useful. In fact, if I had my spectrum analyser back from the manufacturer I'd check the GPSDO noise right now.

73, Peter.
Logged
Pages: 1 2 3 [4]   Go Up