DIY amp testing (with AES17 filter)

[mcdn]

It's built! And it works in so far as it doesn't create any issues when used in loopback testing. The image below is RTA with a 1KHz signal, using a Scarlett Solo for both output and input. We're getting 94dB SNR, which is pretty close to the 100dB spec on the Solo. I can't do any better with a TRS cable between output and input, so we are at the limits of the device.

Switching the diodes in to limit output to 8V P-P drops SNR by 5dB, which seems reasonable given the intent of their use.

I haven't got the 'scope out and tested with a class D amp to check the HF noise suppression yet, but there's no reason to assume it won't work.

The full KiCAD project, BOM etc are at https://github.com/mattclarkdotnet/aes17_filter_25

 

[IVX]

THD+N -94db isn't bad at all for a class D amp.

 

[mcdn]

Here's the Solo measured with just a TRS cable between output and input. It's actually worse - I suspect coupling internally.

 

[mcdn]

THD+N -94db isn't bad at all for a class D amp.

This is just the filter in between the output and input of a Scarlett Solo. Amp testing comes next

 

[mcdn]

Here's a test of the filter effectiveness. I don't have a laboratory source of wideband ultrasonic noise, so this is the output of a TPA3255 amp module driven at ~5W power into a dummy 4R7 load from a bench power supply at 36V. The scope is a Picoscope 2204A, which has only 8 bit resolution. But it has enough bandwidth, and the noise we are looking for is significant, so...

Without filter we see the expected switching noise peak and sidebands at 590 and 1800KHz, plus lots of other "grass". With the filter engaged we see the expected large reduction in all of those. There is still a bit going on but not unexpected given the module is not in a chassis, so it's radiating EMI everywhere and the probe is AC coupled. I'd imagine a higher resolution scope with a differential probe would show cleaner results.

Does this seem OK? If so I will conduct some actual amp testing over the weekend.

 

[St66666]

Here's a test of the filter effectiveness. I don't have a laboratory source of wideband ultrasonic noise, so this is the output of a TPA3255 amp module driven at ~5W power into a dummy 4R7 load from a bench power supply at 36V. The scope is a Picoscope 2204A, which has only 8 bit resolution. But it has enough bandwidth, and the noise we are looking for is significant, so...

Without filter we see the expected switching noise peak and sidebands at 590 and 1800KHz, plus lots of other "grass". With the filter engaged we see the expected large reduction in all of those. There is still a bit going on but not unexpected given the module is not in a chassis, so it's radiating EMI everywhere and the probe is AC coupled. I'd imagine a higher resolution scope with a differential probe would show cleaner results.

Does this seem OK? If so I will conduct some actual amp testing over the weekend.



View attachment 316966View attachment 316967

Built one of these and tested it on a HP3585A audio analyzer. Old, but still good.
Connected the output of the filter to the 50 ohm input of the analyzer and grounded one side of the filter both on the input and output.
This is what I got.
The lowest dip is about 45dB at 248KHz.
Pretty close ...

 

[mcdn]

Built one of these and tested it on a HP3585A audio analyzer. Old, but still good.
Connected the output of the filter to the 50 ohm input of the analyzer and grounded one side of the filter both on the input and output.
This is what I got.
The lowest dip is about 45dB at 248KHz.
Pretty close ...

Awesome, thank you! I hardly dared imagine someone would build one so soon! When you put it in a case, don't add the diodes. There's an updated design without them on GitHub, which also simplifies the jumpering/switching for the resistor dividers.

Are you sure you mean 50ohms on the input of the analyser (output load for the filter)? That should cause a nasty 12dB peak around 30kHz which isn't there on the trace. The design really only works for loads 10k and above (and preferably 50k).

I have better test kit now so can test it with signals up to 176kHz. Will report back.

 

[St66666]

Awesome, thank you! I hardly dared imagine someone would build one so soon! When you put it in a case, don't add the diodes. There's an updated design without them on GitHub, which also simplifies the jumpering/switching for the resistor dividers.

Are you sure you mean 50ohms on the input of the analyser (output load for the filter)? That should cause a nasty 12dB peak around 30kHz which isn't there on the trace. The design really only works for loads 10k and above (and preferably 50k).

I have better test kit now so can test it with signals up to 176kHz. Will report back.

Was looking for something to build and test. Perhaps improve on if needed.
Appreciate your effort and it made it much more easy for me. Would have maybe chosen different coils, but it was convenient to just copy what you made and test it.
The analyzer can go from 20Hz to 40MHz and has 3 options on the input. (that would be the load on the output of the filter)
Either 1Mohm or I can use a different input where I have two options 50 ohms or 75 ohms.
Don't see a huge difference beside the whole curve dropping lower when using the 75 or 50 ohm input.
Don't see any peek around 30 KHz either. Maybe because I connected one side of the input to the coax ground and connected that to the filter ground? Plus I don't have the 4 ohm Rdut in there just the filter.
If I remove that ground connection I get a hump around 153KHz and two dips at 122KHz and 215KHz.
Included another sweep from 20Hz to 2MHz using the 75ohm input

 

[mcdn]

Here's the response I get up to 174kHz. A bit more drop at 20kHz than ideal [EDIT: see later measurement into 100k load], but the overall shape is good.

 

[sarieri]

Here's the response I get up to 174kHz. A bit more drop at 20kHz than ideal, but the overall shape is good.View attachment 327032

That’s amazing. Did you measure if there’s a SINAD loss from this filter?

 

[mcdn]

That’s amazing. Did you measure if there’s a SINAD loss from this filter?

Using a 1KHz tone from the Topping D10 Balanced, direct to the Cosmos ADC I get 117dB SINAD, which is basically the same as Amir measured for the D10 Balanced (he got 118dB). Adding the filter drops that to 114dB, but part of that will be insertion loss because the Cosmos ADC has a low input impedance, and I can't get my Cosmos Scaler working properly due to a grounding issue somewhere. Call it 3dB worst case.

The AP AUX-0025 is specified at <110dB distortion, so we are in the right ballpark.

 

[sarieri]

Using a 1KHz tone from the Topping D10 Balanced, direct to the Cosmos ADC I get 117dB SINAD, which is basically the same as Amir measured for the D10 Balanced (he got 118dB). Adding the filter drops that to 114dB, but part of that will be insertion loss because the Cosmos ADC has a low input impedance, and I can't get my Cosmos Scaler working properly due to a grounding issue somewhere. Call it 3dB worst case.

The AP AUX-0025 is specified at <110dB distortion, so we are in the right ballpark.

Thank you for the info. Well, I’ve been struggling with APU’s ground noise problem lately. Scalar has been working fine…

Anyway, I think 114dB is already very good since best measured class D like NCx500 only measures 113dB SINAD. Maybe adding a scalar or shield the inductance will give better results.

 

[mcdn]

Here's a better measurement, using the Cosmos Scaler as the load - the previous measurement was direct into the ADCiso which has a low input impedance. I fixed my grounding issues by powering both the scaler and the ADCiso from a power bank, with the D10 powered from the laptop. The FR drop at 20kHz is now <0.5dB. Distortion remains around -114dB.

 

[amirm]

Great effort you all!

FYI I use the AUX-40 which is flat to 40 kHz. As well built as typical AP products are, the PCB traces in one channel smoked out! Have no idea how that is possible but it happened.

 

[mcdn]

Great effort you all!

FYI I use the AUX-40 which is flat to 40 kHz. As well built as typical AP products are, the PCB traces in one channel smoked out! Have no idea how that is possible but it happened.

Happy to build you one of these if you want

 

[amirm]

Thanks. I was going to build my own but then someone told me my measurements would always be subject to doubt due this box not being an official AP. So bit the bullet and bought it even though the cost is just not justifiable.

 

[St66666]

Here's a better measurement, using the Cosmos Scaler as the load - the previous measurement was direct into the ADCiso which has a low input impedance. I fixed my grounding issues by powering both the scaler and the ADCiso from a power bank, with the D10 powered from the laptop. The FR drop at 20kHz is now <0.5dB. Distortion remains around -114dB.View attachment 327693

If I may ask, which software are you using?

 

[mcdn]

If I may ask, which software are you using?

REW

 

[mcdn]

To answer the question on what impact adding the filter has on your measurements, here are the 1KHz FFTs without filter (1st) and with filter (2nd). There's no appreciable difference, we're just seeing some limitations of my test setup as the noise floor should be a bit lower below 1kHz. Still -120dB A weighted isn't bad.

 

[sarieri]

So I made one as well, with better caps. And it measures pretty good. Setup is D6S—AUX0025–Scalar—APU—ADC. First graph is without AUX0025, second one with it. The noises at low frequency is probably EMI or grounding. Those noises vary by the position I put my measurement gears.


Btw, I did a sweep as well. -0.25dB at 20kHz. I measured this by feeding D6S+LPF directly into ADC. Probably gonna get better result if use scalar.