Osc + Notch Rev D

[matt]

This documents some follow-up to the last post on this topic (here). REVD of the hardware is back and has been built. There were two big changes made: The first was to restructure the lineup so that the 1 kHz oscillator was followed by a 0/12 dB gain stage, and that is followed by the attenuator. That is reflected in the app/block diagram below:

The second big effort was to completely shield the different sections. The photo below shows a board without shields installed.

With shields installed, the board appears as follows:

The shields were laser cut from 0.5mm 0.02" Nickel Silver (H02 1/2 hard) from OshCut. They are a bit heavy and too expensive for production, but solderability is fantastic. Ideally for shielding below 10 or 20 MHz you’d want something with more steel.

The shields ship flat can be bent into shape:

In the end, the shielding doesn’t seem to have made much of a difference. The 2H might even be a little worse coming out of the notch with the input shorted:

The risky change was moving the amp ahead of the attenuator. This is because it required an extra opamp inserted in the signal path. SPICE suggested it probably wasn’t an issue, and in the end it appears SPICE was correct.

The oscillator center frequency on this build was reported at 996.26 Hz.

Sweeping the notch yields the following. Note that the notch depth is almost 70 dB. This is better than the unit tested on RevC, but not due to any changes. It’s just the luck of the draw given part tolerances.

We can export the notch and get some key metrics on the notch performance. Note that at the fundamental we’re getting 54 dB of attenuation.

With the notch loaded correction loaded into the User Weighting, we can look at the oscillator output through the notch with the correction applied at roughly 6 dBV. Note our output is indicated at 17.17 dBV. This is because the notch has 12 dB of gain built-in to the output stage (see block diagram above)

Normalizing and upping the FFT size to 256K and 5 averages gives a THD figure of nearly -150 dB. Now, it’s not clear why this should be any better or worse than the RevC hardware. The layout is a bit cleaner, and there are shields, but I’d be inclined to believe it’s more part tolerances than anything else. But more study would be needed.

And engaging the +12 dB amp on the oscillator output yields the best-case THD+N

That’s all for now!

[1audio]

Great work. The one issue I got bit by some years ago was oscillator drift. It moved enough that the harmonics move almost a bin and were significantly attenuated. Later I discovered that it was not really that good. Can you source lock your oscillator?

[matt]

Hi @1audio, yes, the oscillator can be injection locked, but the tuning range is very, very small and the harmonics degrade. A few 10ths of a hertz either way until you are just left with beat frequencies. (EDN link for those wishing to learn more)

The drift issue is indeed important and hard to know with 100% certainty, but here’s my thought process: On the QA401 the reported peak is the real peak, and not an RMS measurement around a range. Take a look at a peak measurement in loopback of a very small 1K FFT. Note that each bin is 46.8 Hz, and the generator has been adjusted to be in the middle of the bin (984.375 Hz). Peak is -10.04 dBV

If I move up half a bin width (23.4 Hz), then the peak drops from -10 dBV to -11.46 dBV and you can see the peak has a blunt top (straddling two bins). So, if you see a peak amplitude change that will signal you are wandering outside of a single bin. But even if you straddle two bins perfectly, your drop in level will be about 1.5 dB. Now, if you are wondering across 4 or 5 bins, then it’s more of a spread spectrum action on the harmonics, and they could artificially reduce by a multiple of the 1.5 dB seen when perfectly straddling two bins.

Next, I verified the stability of the oscillator with the wow and flutter tool. This tool looks at interpolated zero crossing to compute instantaneous frequency. At 256K FFT, each bin is 0.183 Hz wide. I’ve superimposed the bin center and the bin width (red box) on the FFT, mostly just for scale. You can see there is some wandering, but it’s relative to the bin width it’s not bad. BTW, the oscillator getting so close to bin center is just luck.

I’m sending the box to a buddy from my days at Motorola for more measurements on some different equipment and will report those when they come back. And if you have the time, I could send it to you too for eval when it comes back. It’s all self contained: Just plug in the USB, let it warm up, install the UI and start measuring.

[1audio]

Those are the same issues that arose when I tried injection locking. I guess a real PLL would be needed to properly lock an oscillator. However the excellent results you are getting from the optocoupler suggests they may work in that application as well. The real downside is the slow response they have.

I would be delighted to test and compare your box with the others I have here.

[matt]

OK, great! It’s hopefully headed to Europe this week for someone to test, and when it comes back I’ll send it on.

[RNMarsh]

Did I miss it… are you going to mfr and sell this as product? I would buy one.

THx, Richard N. Marsh

[matt]

Hi @RnMarsh, no plans for a product as I don’t think there’d be enough volume. But I think it’s worthwhile to have a free loaner unit that can be sent around for people to validate their own setups if needed. I certainly don’t think it’s the best performing unit in the world, but with the USB plug-and-play and built-in attenuator I think the really-good-performance combined with ease of use can help a lot of folks gain confidence in their bench setups.

[BKDad]

Matt,

Dunno if the product would have legs or not - that’s up to you. But, there’s very likely a small but serious group who’d be willing to build such a device for their own use. All it would take is a schematic, BOM, and PCB layout to make the job easy. Just a suggestion.

Clarke

[matt]

Hi Clarke and @RNMarsh (from the other thread). I think we’ll build 25 of these. One of these is now with a consultant in Germany who has made some measurements on an apx555 with the analog options. He’s found same issue as above, where the notch is contaminated with 1H and 2H from the oscillator. I’ll do another PCB to address (since the shields showed it wasn’t radiated). I think the solution is to run the 1 kHz and notch each from their own LDO.

In any case, if interested in purchasing one once the performance is improved to -150 dB THD then please send a note to sales (at quantasylum dot com) with the subject “QA480” so we can gauge interest. No obligation. At this point, I think they’ll be $199 to $249 including a case (front panel on the current units is below). Just plug it into USB and go. +18 dBV max output, 0…63 dB attenuator, notch with 12 dB of gain (probably moving to 20 dB) and switched bypass on the notch.

I think this will let a QA401 measure a state of the art DAC or pre-amp at 1 kHz that previously only a $30K analyzer could measure with margin.

[RNMarsh]

Hi Matt,

yes, I will buy one, for sure.

Thank you,
Richard Marsh

[1audio]

With some simple changes it would be more useful, Add an output attenuator (pot) and an input attenuator so it can take the output of an amp or other higher level signal. I’m definitely interested. I can see using it as the basis for an OPAMP tester for example.

[ilya]

I’d be interested in this unit as well (hopefully the audio field jobs here recover enough so I can afford it)

[matt]

Follow-up post is here: Introducing the QA480

[RNMarsh]

Put me on the mail list to buy one when ready.

I can measure to -160 re 1 v/ 1khz now but not in anything so small and portable.

Thx,
RNMarsh
(Richard)

[matt]

Got it, thanks! The final PCB was released last week, and the PCB build should be the week of August 17. Then a week of logistics. Hopefully they are all sent out by end of August.

In roughly two weeks (mid August), I’ll send a link to folks to purchase that have sent an email to sales with “QA480” in subject line. And whatever is left after that will then go on the website for order as a normal product.

Thanks!