Pages: 1 2 [3] 4   Go Down

Author Topic: Baseband FFT system - any h/w and s/w experience?  (Read 704 times)

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #30 on: October 13, 2021, 03:18:35 PM »

I couldn't resist trying to design an SSB filter with these crystals and see below for a 6 pole filter simulation using the s-parameter data.



This is only a 6 pole filter. There is a strange blip in the lower stopband, this might be an issue with the interpolation between points or it might be a real artefact of the crystal.

An 8 pole filter would probably have very good performance. The 6 pole filter is about 3kHz wide and only has a loss of about 1.3dB and has very low passband ripple. This assumes all 6 crystals out of the bag are the same of course...

Edit:  I just looked on a wider span and this crystal does have some nearby resonance modes that would spoil the performance as an SSB filter. However, I think it could still be used as a filter in a phase noise measuring system.
« Last Edit: October 13, 2021, 03:38:12 PM by G0HZU »
Logged
Regards
Jeremy

VK6HP

  • Member
  • Posts: 1222
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #31 on: October 14, 2021, 11:55:41 PM »

Hi Jeremy,

That filter looks pretty good, notwithstanding the spurious responses you mentioned in the addendum.  Probably quite usable.  You can't argue with the price, either: AUD 3.27 for a bag of 10.  Sifting through some of the other offerings I was looking to see if there were any better-quality crystals that would support direct carrier notching within a ham band, while still being useful in a general-purpose heterodyne system.  Probably understandably, I didn't see any, although there are certainly better looking crystals with low ESR.  For example, this 11.0592 MHz 35 ohm ESR type (https://au.rs-online.com/web/p/crystal-units/2261752) would still be under AUD 10 for 5 pieces and would, I think, work well in the notch filter test box I've used for the MF crystals (200 ohm transformers and appropriate synthetic quarter-wave lines). 

73, Peter.
Logged

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #32 on: October 15, 2021, 02:23:17 PM »

Thanks! I had another play on the simulator this evening and designed an 8 pole version and I've done it with standard (fixed) cap values. It looks very promising.

See the simulation below. I also think the spurious terms seen on the VNA aren't going to be as big an issue as I thought. I would hope to be able to get 100dB stopbands on a real filter if well grounded and shielded etc. Hopefully the stopbands will extend out at least +/- 1MHz but I won't know until I build it. I have done this stuff before so I think the results should be quite good.



This design went together quite easily this evening. However, I think I need to take another set of s-parameters from the crystal because I need finer resolution to really get the passband as flat as possible with decent group delay. It already looks very good but I think it's worth refining it as there isn't anything to tweak as it all uses fixed caps. So it is a case of build it and hope it is the same as the simulation. It should be but the capacitors need to be good quality with tight tolerances.

I haven't really thought which HF bands this crystal filter would suit but a pack of ten crystals would be enough for the SSB filter and the product detector.

Logged
Regards
Jeremy

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #33 on: October 15, 2021, 02:37:37 PM »

I've also just tweaked it on the simulator for a narrow SSB version and this has just under 2kHz bandwidth and it looks just as nice. I've not tried for a cw version but it might be possible to get the bandwidth down to 1kHz.
Logged
Regards
Jeremy

VK6HP

  • Member
  • Posts: 1222
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #34 on: October 17, 2021, 03:37:13 AM »

It's looking pretty good Jeremy.  Along similar lines and as a bit of a sideline to my original enquiry I've ordered some 10 MHz and 11.06 MHz CTS crystals (https://au.mouser.com/ProductDetail/cts/mp100-e/?qs=tjlMjqRIEYRDVgTydYa6ig%3D%3D&countrycode=AU&currencycode=AUD) just to see how they go in a carrier notch filter.  I figured that 10 MHz is useful enough on its own but the "odd" frequency might also have value in avoiding common harmonics etc.  There's not a lot of product detail for the purist RF designer but I guess we'll see how good the crystals are during measurement and prototyping.

With the SML-01 as an LO in a mix-up or mix-down system and a low-noise IF amplifier, there are a few useful experiments to be done in seeing how far I can push my RSA3015E-TG spectrum analyser.  I can't recall where I saw the idea but it's probably possible to simply parallel the 10 and 11.06 MHz crystals in a multi-resonator filter to make a single box with a dual-notch.  Much depends on the crystals and the placement of spurious resonances, I guess.

73, Peter.

« Last Edit: October 17, 2021, 03:43:51 AM by VK6HP »
Logged

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #35 on: October 17, 2021, 06:09:55 AM »

I've not tried to make a decent notch filter with a crystal filter. I just had a quick go using the s-parameter file for the 9.83MHz crystals and got the result below.

This gave a notch depth of 55dB but the notch is 1kHz wide at the top. I'm not sure if there are ways to make it tighter at the top. The plot below is a simulation using 5 crystals.



This required a new filter topology that I've not tried before. I think this would be more interesting to build than the SSB crystal filter just to see if it will work as per the simulation.
Logged
Regards
Jeremy

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #36 on: October 17, 2021, 09:32:48 AM »

I just built the notch filter and the response is exactly as the simulation with 55dB depth to the notch and the same width at the top.

I tried it on a few sig gens that I have here at 9.828MHz and 0dBm carrier power. The results were quite revealing. At first I though something was wrong somewhere as the Agilent ESGD sig gens both struggled to better -135dBc/Hz at wide offsets. The Mi2024 was a little better. The Marconi 2019 and 2022C were much better and managed about -147dBc/Hz at wide offsets. I set the sig gens to +13dBm power level with a 13dB external attenuator.

I think the notch filter could benefit from about another 10dB notch depth but apart from that it works extremely well.

Note that when I built the filter I cherry picked the crystals that had the same notch frequency. There was a spread of +/- 150Hz so I think I will need to do the same if I try and build the SSB filter. Otherwise it may not give a flat response.

It would probably be wise to buy 20 crystals (total cost £4) to select 8 matched crystals to make up the 8 pole SSB filter.

« Last Edit: October 17, 2021, 09:37:30 AM by G0HZU »
Logged
Regards
Jeremy

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #37 on: October 17, 2021, 09:46:07 AM »

I guess the hunt is on to buy/make a high level mixer and a versatile low phase noise generator. I have a few DDS eval boards here that might be OK as the LO and I have quite a few wideband level 13 mixers like the MiniCircuits ADE42MH.

However, I think a level 17 mixer might be better. I'm not sure how much power the notch filter can deal with but the level 17 mixer might handle +7dBm RF and this would put 0dBm into the notch filter at the IF port.

Logged
Regards
Jeremy

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #38 on: October 17, 2021, 11:23:19 AM »

I tried the notch filter on an LC oscillator at 9.828MHz and I used a different spectrum analyser this time. This analyser has a DANL of -169dBm/Hz with the preamp on and I managed to see -161dBc/Hz at 40kHz offset using the notch filter to null out the carrier. This is 20dB better than I could measure before using the Tek analyser directly.

However, the LC oscillator was sometimes being pulled by the notch filter and this made it difficult to fully null the carrier. I'll try a crystal oscillator at 9.828MHz next.

I might also try the Analog discovery 2 to see how it copes here as the analyser. It will need a preamp I think. The software is a bit clunky so it probably won't work as well as the hardware suggests it could. It has a 14bit ADC sampling at 100MSa/s.
Logged
Regards
Jeremy

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #39 on: October 17, 2021, 01:16:29 PM »

I put together a really quick and dirty MMIC based crystal oscillator using one of the 9.83MHz crystals. This oscillates on the series resonance of the crystal so I was able to find a crystal amongst the rest I have that oscillates closest to 9.8279MHz.

This gave the following results when using the notch filter:

1kHz offset  -156dBc/Hz
2kHz offset  -159dBc/Hz
10kHz offset  -163dBc/Hz
100kHz offset  -163dBc/Hz

This is fairly close to my earlier prediction for this crystal using Leeson's equation in post #29. However, this oscillator seems to be slightly noisier close in and this may be because of the flicker corner frequency of the MMIC. Otherwise, the ultimate noise floor at -163dBc/Hz was spot on with post #29.

There's probably an element of luck in this as I'm driving the resonator closer to 0dBm than -6dBm and the MMIC noise figure is quite high but it's good to see the result showed very low phase noise at >=1kHz offsets. The phase noise floor is probably closer to -165dBc/Hz because the analyser noise will be affecting the result a bit here. This is then lower than the prediction in post #29 but this is probably because I'm driving the resonator with slightly more power than that used in the spreadsheet.

Either way, it's not bad for a first attempt at a low phase noise xtal oscillator using a £0.20 crystal...



Logged
Regards
Jeremy

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #40 on: October 17, 2021, 01:50:48 PM »

I've now added another crystal section to the notch filter to get a notch depth of nearly 65dB. The notch filter now has 6 crystals.

Here's three phase noise plots using the notch filter at 9.828MHz. The top trace (pink) is the phase noise of an Agilent E4433B vector sig gen. This has really poor phase noise on the HF bands as you can see in the plot below. It's approx -103dBc/Hz across the whole 10kHz span.

The cyan trace is the phase noise of one of my Marconi 2019 sig gens. This sig gen has quite low phase noise and shows about -141dBc/Hz at an offset of 4.5kHz.

The yellow trace at the bottom is the MMIC/crystal oscillator and this shows -163dBc/Hz at 4.5kHz offset. This oscillator doesn't quite sit in the -65dBc null of the notch filter but it is low enough to not upset the spectrum analyser.

This crystal oscillator has 60dB lower phase noise (at 4.5kHz offset) compared to the E4433B sig gen!

Logged
Regards
Jeremy

HAMHOCK75

  • Member
  • Posts: 1297
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #41 on: October 17, 2021, 02:01:58 PM »

Quote from: G0HZU
However, this oscillator seems to be slightly noisier close in and this may be because of the flicker corner frequency of the MMIC.

I found in the past that device manufacturers were not very accurate in their characterization of the en, in noise of devices at low frequency where the device starts showing 1/f  behaviour. I used an HP4470A transistor test set to characterize devices rather than rely on datasheets.
Logged

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #42 on: October 17, 2021, 02:33:48 PM »

Quote from: G0HZU
However, this oscillator seems to be slightly noisier close in and this may be because of the flicker corner frequency of the MMIC.

I found in the past that device manufacturers were not very accurate in their characterization of the en, in noise of devices at low frequency where the device starts showing 1/f  behaviour. I used an HP4470A transistor test set to characterize devices rather than rely on datasheets.
Thanks. I'm not really sure what the noise contribution will be from the MMIC at 1kHz offset. It's far from an ideal device because it also has a 5-6dB noise figure. I may also be driving the crystal too hard and this may cause issues with the circuit Q.

The MMIC oscillator was just a convenient circuit to use quickly. It is my old MMIC based LC oscillator I once designed to demonstrate Leeson's equation. I swapped the LC tank for the crystal and dialled in some attenuation in the loop. A simulation predicted the phase noise plot I gave back in post #29. It's only in error by 3 to 4dB but I was expecting it to be closer to -160dBc/Hz at 1kHz offset.

I'll try again with a BJT and maybe also a JFET  to see if I can get closer to -160dBc/Hz at 1kHz offset. I'm not sure the company's Agilent E5052A SSA can measure phase noise this low at this offset. It might be able to using correlation but it can only accept inputs at 10MHz or greater. So I can't measure this oscillator at work. The E5052B can accept baseband inputs but our older E5052A can't do this.

I'm due to use the company E5052A again in a few weeks' time so this would be a good time to do some comparisons with oscillators >=10MHz.

Logged
Regards
Jeremy

HAMHOCK75

  • Member
  • Posts: 1297
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #43 on: October 17, 2021, 02:52:02 PM »

Thank you for posting the IEEE article on the low phase noise crystal oscillator. A very interesting read.

There is a commercial unit that has almost identical close in noise.

https://wenzel.com/model/btuln/
Logged

G0HZU

  • Member
  • Posts: 345
Re: Baseband FFT system - any h/w and s/w experience?
« Reply #44 on: October 17, 2021, 04:32:08 PM »

Yes, Charles Wenzel is a clever chap. It's no surprise to see such good numbers for an oscillator. I'm still not sure how they do it. I could probably improve the MMIC oscillator by amplifying it and then running it through the SSB crystal filter. This would give it good numbers at 10kHz offset as long as the filter could handle the drive power.

However, I just connected it to the Tek baseband analyser and the phase noise at 10Hz offset was poor at -90dBc/Hz. This was much worse than my simulation in post #29. By dialling in more attenuation in the feedback loop and by reducing the MMIC bias current I managed to get it down to -100dBc/Hz at 10Hz. The phase noise at 100Hz offset was nearly 30dB better and this indicates that there is a lot of flicker noise present (30dB/decade rolloff rate). It is still worse than expected though.

The IEEE article really is informative and well written. It would be nice to get hold of the matched BJTs used in that article. I have some low noise BJTs here but I'm not sure of the flicker noise performance.
Logged
Regards
Jeremy
Pages: 1 2 [3] 4   Go Up