Waiting for QA480 so Building a Notch Filter - Bainter vs Twin-t

@matt or anyone else here, while waiting for QA480’s to be back in stock, I built a passive twin-t with a notch depth of 53.2 @1007.8 hz. I have been considering doing a bainter notch to get a higher Q and avoid sensitivity to component values. Any thoughts on this noise wise?
Also built a simple amp with a 5532 to follow the twin-t and have played with gains from around 12-40, but noise levels are too high. Anyone have a good low noise design for something to follow the notch filter?

Hi @Moto, in my experience you don’t need a lot of notch rejection, as the THD gets really good at -30 dBFS or so. And the higher Q just results in more tuning and drift headaches. The QA480 used a twin-T with a fixed Q, and 12 dB of gain post notch.

I’ve also built a lower-noise notch for some other experiments. Note the R has gone from 3.57K to 732 and the C has grown to 220n. This is about the largest NP0 C you can get (it’s an 1812 package). Also, the opamp moved to OPA1612 for noise and gain was increased. The circuit below shows about -116 dBV equivalent input noise, which is plenty of margin to measure some of the modern class D amps (like Purifi) that are basically free-running oscillators.

Thx @matt! So I breadboarded the twin-t and ended up with about a 55db notch @1005.6hz. With the QA402 Gen at 0dbv out I end up with 3 or so mv output from the notch.
The only opamp that was fairly low noise that I had on hand was the NE5532. So I understand why the ratio of resistors in the opamp circuit to get 18db gain, but 2 questions.

  1. Why 18db gain choice?
  2. Why that specific absolute value for those resistors( except of course the ratio)?
    Sorry if I’m bring ignorant but this is really a rekindled hobby at 73 years old from my very distant youth.

Hi @Moto, congrats on the success!

Some quick definitions: Circuit A above is the QA480 notch, and it has 12 dB of gain. Circuit B is an experiment in very low noise diff to singe-ended conversion with a notch, striving to get >100 dB CMRR. The aim here is something you where you can connect floating scope probes across a free-running oscillator type Class D output and accurately capture the signal of interest.

The QA480 (Circuit A) has 12 dB of gain on the notch, the second notch has 18 dB. Both are somewhat arbitrary. The QA401 and QA402 analyzers have a noise floor of -115 dBV (20 kHz, no weighting) and the aim is to get the output noise of the QA480 suitably above the input noise of the QA40x. On the QA480, if you short the input of the notch, then the output noise measure about -106 (20K, NW), so roughly 10 dB above the QA40x noise floor. And this means input referred, the input noise into the notch is about -118

Circuit B, the second notch (with 18 dB) moved to a differential input, OPA1612 versus OPA1656 followed by a precision diff to single ended conversion (discrete instrumentation opamp, with matched 0.01% resistors used for diff to SE), an attenuator, a 2nd order low pass filter, and then the low-noise notch you see above (the one with 18 dB). SPICE was used to inform the decisions. That notch has -98 dBV output noise, which, with 18 dB gain gives an input noise of 116 dBV (20K, NW). So, a little worse. But it’s doing a bit more work. Still, it has plenty of margin to measure the new class D.

In both cases, additional gain can be had easily, of course, but if you look at opamp THDN curves, they are usually happier with lesser gain. So, just enough, but not too much. If you have a choice in your next opamps, give the OPA1612 and OPA1656 a try. If you’ve gone with higher valued resistors (closer to Circuit A), the OPA1656 might do better as the OPA1612 noise degrades quickly when you see 1-2K or higher resistors around it.

Thx @matt. Yes the NE5532 is way too noisy. I will get some 1612’s and 1656’s. Looking at the second schematic with the 18db gain, using a 0dbv input at the front of the notch, what vrms output level would you see from the opamp?

Hi @Moto, the notch depth should be about the same, you’d just have 6 dB more gain so the 1 kHz would appear 6 dB higher in the second versus first schematic.

Sorry @matt I was not being clear. Not considering the first circuit at all, with 0dbv input to the notch on the 2nd circuit, what did you see as vrms output from the circuit?

Hi @moto, the notch shape (“Q”) should be the same for both circuits all considered. Circuit 2 just has 6 dB more gain, that’s all.

@matt so on circuit 2 you get say -55db + 18db = -37db out right? With 1vrms in then that’s 14mv into the QA402. How is the adc on the QA492 with 14mv in?

@matt I rebuilt just the passive notch of circuit 2. This is the weighting file. I will use the Gen for input so set filter to exact center of the notch.

Setting Gen1 to 1006.71 I get the following results through the notch.

  1. Not sure why RMS reads 2.6 dbr. It was closer to 0 before.
    Is this causing the extraordinarily good readings?
  2. Is it the signal from GEN1 causing the noise level to be higher and being depressed by the notch?

@matt, I have a question about your blog post on creating a notch filter. You suggest doing the sweep at 0dbv which I did with the following result.

However, my oscillator has its best performance at 7.7dbv. When put through the notch, here is the result.

Do I need to recharacterize the notch at 7.7dbv to use it with the output of my oscillator at 7.7dbv?

@matt, could you give me your view on this? By the way your suggestion of switching to the opa1612 was like night and day.

Hi @Moto, you shouldn’t need to re-characterize for a higher level as long as the measurements are all staying out of the noise (and it looks like they are)

The 2.6 dBr means you have some error in the setup. If the notch is really deep, then the oscillator can ride up and down the skirt, and it will manifest as amplitude “breathing” and it can be very hard to get everything settled down. There’s a benefit to a notch that is “good enough” but “not too good”. But it’s hard to quantify precisely.

Thx @matt. Changed some cables out and got this.
The oscillator is an smsl su-9 as you can see is set to 1005 to get it off the bottom of the notch. Now rms dbv showing as 6.50 when the output of the dac unloaded is 7.61.
The measured gain of the opa1612 is 18.4 and has been entered as such (I used slightly different resistor values than you). Do I just ignore this?

@matt I built a new notch following your suggestions and tweaked until I go this as repeatable.

The best and mist consistent results were gotten off the notch bottom using 1010hz.

With that I use REW to generate the sine at 1010 and the passive only notch delivers.

The SMSL SU-9 dies not seem to be having a large noise impact re the notch( am I interpreting that correctly?).

I did not use an input buffer opamp but built a 12db gain amp using your schematic above with an opa1612.

This however didn’t improve things.

The noise seems to be affected by the notch now. Do I need to simply kerp working on the power supply and 12db gain amp to reduce the noise and added harmonics?

Thx for any help.

Hi @Moto, it’s hard to know where to go next. For the picture below, you used REW driving a DAC…are you sure the DAC isn’t contributing and that the harmonics aren’t real? Also, are you sure the caps used in the notch aren’t contributing? They should be NP0 or C0G, but there can be differences among cap makers too, even if both are using the same dielectric category.


@matt, I was using an op amp as a rail splitter from a 12v battery. I then tried just 2 9v batteries and got the result below so it appears the problem is the op amp and related parts in the rail splitter. These results on the smsl su-9 look similar to those on the audioscience review.
Any suggestions for a low noise rail splitter circuit off a 12v battery?

Above was me touching a chassis. This a bit better re thd.

Hi @Moto, looks like you are pretty much there! Your THDN is -120.3 dB, and your N-D (noise minus the distortion) is -114.79 and your peak is 6.82 dBV. So, your noise level is 114.79 + 6.82 = -121.6 dBc, and your harmonics are -129.46 dBc, or about 7.9 dB better. So, the degradation from -121.6 (noise dBc) to the measured -120.3 (n+d dBc) is probably coming from the harmonics. But, your harmonics are no longer the major contributor–the noise is now driving THDN. Now, it’d be nice if you could get another 10 dB out of the harmonics…the only thing I can offer there is tweak, tweak some more and then tweak some more. And I think you are deep into that already!

Your change in supply topology is offering a major clue–the system looks very sensitive to changes there.
What has always looked like a solid project to me is Jan Didden’s Silent Switcher. I met him a few years back at a Rocky Mountain Audio Fest (a few years before they moved to the new hotel) as I was loading up on back issues of Linear Audio. Maybe this would meet your needs?

Thx@matt. I was able to achieve the same results yesterday with a 12v iron phosphate battery and dramatically simplifying my virtual ground to 2x100k resistors as a voltage divider and a 50 ohm resistor on the output of a lm4562. I’m fine for now using the 12v battery with this virtual ground but will look into that kit. My measurement of the smsl su-9 are basically the same as audioscience so probably not much more to be gained with this device.
Any news on ordering the new versions of product you mentioned?