Make a good measurement system with REW and Raspberry

[Antonio Di Motta]

Ok, let me know how the project is progressing and the results you get.
Each experience is very interesting.

Greetings

 

[JLM1948]

Ok, let me know how the project is progressing and the results you get.
Each experience is very interesting.

Greetings

please see attached FR and NS
For some reason REW does not want to sweep beyond 80kHz, but I spot-checked the response, -3dB at 82kHz, -6 at 84 and -10 at 85kHz.
I could probably get a better response by thoroughfully selecting capacitors. For example I needed a 4.7pF, that I did not have so I used two 10pF in series, but they are 10% tolerance, and I have no way to measure them with enough accuracy. I have a cap tester that consistently gives a lower reading with small caps. Typically a 47pF reads at about 40.
The noise increase at about 78kHz is due to one of the filter stages that has a very high Q (= very high noise gain).
Anyway, it makes the system a little more usable, although the remaining noise level is enough to fool the rms detector in the HF band, that has a BW of 160kHz. It's the price to pay for having less than 0.1dB loss at 20kHz. It is a problem only for testing, because in actual operation, the signal chain is strictly restricted by higher order filters.

 

[Antonio Di Motta]

I hope you have calibrated the sound card first and only then plotted the overall response curve including the calibration corrective curve As for the maximum frequency, you can set this in the measurement mask in start and stop frequency, perhaps the stop frequency was set to 80Khz. However you can set it up to 96Khz for 192Khz sampling rate.

 

[Antonio Di Motta]

I would have expected a loop out / in sound card calibration curve like this one I'm attaching to you which is actually my 192Khz sound card.
Consider that my band is currently limited by the output filter to the DAC at about 40Khz, since at the moment I do not need to verify beyond this frequency.
Once the calibration is done, the response measurement must be perfectly compensated in the selected frequency range and you should not have that peak at 80Khz which is quite anomalous.

 

[JLM1948]

I hope you have calibrated the sound card first and only then plotted the overall response curve including the calibration corrective curve.

Why would I need a corrective curve when the response is flat to 92kHz, without filter?
[/QUOTE] As for the maximum frequency, you can set this in the measurement mask in start and stop frequency, perhaps the stop frequency was set to 80Khz. However you can set it up to 96Khz for 192Khz sampling rate.
[/QUOTE] Try as I might, I couldn't enter more than 80kHz in the stepped sine mask.

 

[JLM1948]

I would have expected a loop out / in sound card calibration curve like this one I'm attaching to you which is actually my 192Khz sound card.
Consider that my band is currently limited by the output filter to the DAC at about 40Khz, since at the moment I do not need to verify beyond this frequency.
Once the calibration is done, the response measurement must be perfectly compensated in the selected frequency range and you should not have that peak at 80Khz which is quite anomalous.

This graph shows exactly how much error is produced by the filter. If I had compensated it, I wouldn't have seen this "anomalous" peak. In order to try to fix it, I need to see the uncompensated graph.

 

[Antonio Di Motta]

To carry out all the measurements in REW, the calibration phase is fundamental. If you want to visualize the response of a filter or any quadripole you must first normalize the response of the sound card by looping between input and output. Then you can connect an external device between the output and the input, a filter or an amplifier or whatever, and only at that point are you sure that the response you get is actually the only one of the device as that of your measurement system is been compensated.
This technique allows you to always obtain a faithful response of the quadripole under measurement almost independently of the band of the sound card.

 

[Antonio Di Motta]

Non solo se hai una scheda audio non molto lineare con banda passante 20 -40K a -10db effettuando la calibrazione pui sfruttarla su tutta la banda come se fosse instrinsecamente lineare, cioè come se avesse una banda 20-40Khz 0db.

 

[Antonio Di Motta]

Not only if you have a not very linear sound card with 20 -40K at -10db bandwidth, by performing the calibration you can use it on the whole band as if it were intrinsically linear, that is, as if it had a 20-40Khz 0db band.

 

[Antonio Di Motta]

If you want to be sure of what you are measuring I advise you to always calibrate the sound card, this is my way of operating.
Indeed to say that there are more than one calibration, there is the one relating to the response, the one relating to the impedance, the calibration of the generator levels and the RTA level and also the microphone calibration file.
All these calibrations serve to give reliable indications to REW which behaves as and even better a measuring instrument.
In fact, even the measuring instruments must be calibrated.

 

[JLM1948]

If you want to be sure of what you are measuring I advise you to always calibrate the sound card, this is my way of operating.
Indeed to say that there are more than one calibration, there is the one relating to the response, the one relating to the impedance, the calibration of the generator levels and the RTA level and also the microphone calibration file.
All these calibrations serve to give reliable indications to REW which behaves as and even better a measuring instrument.
In fact, even the measuring instruments must be calibrated.

You don't understand what I did. I calibrated the system in loopback, that shows a flat line up to about 90kHz.
The last measurements are those of the added filter.

 

[Antonio Di Motta]

I am attaching a graph that allows you to evaluate the difference between the response curve of the card and the compensated one.
The black dotted curve is the actual response of the sound card as a sum between output and input, the green dotted curve is the compensated phase and the yellow line the compensated response.
Once you have achieved this, insert your nice filter between the output and the input of the sound card and you will see the real response of the filter alone.
Very simple, so with my sound card that has a bandwidth from 1.8Hz to about 38Khz at -3db I can safely test devices from 0 to 40Khz without having any errors due to the non-perfect linearity of the card.

 

[Antonio Di Motta]

Ok, you did what I was telling you.

 

[Antonio Di Motta]

This graph shows exactly how much error is produced by the filter. If I had compensated it, I wouldn't have seen this "anomalous" peak. In order to try to fix it, I need to see the uncompensated graph.

There was a bad interpretation, I was talking about compensation of the sound card, not compensation with the inserted filter.

 

[goginux]

I don't use raspberry pi for measurements, only for the decibel meter. but now I deleted raspbian, put volumio on it, I have sonarworks and presonus audio card for precise measurements, I used raspberry together with audisey denon.
i have atomic pi to,with 10w amplifier on board,see pictures!
sorry for my bad english,i use google translate

 

[John Mulcahy]

Try as I might, I couldn't enter more than 80kHz in the stepped sine mask.

The end frequencies are at one-third octave centres, 80 kHz is the last one-third octave centre frequency that is less than 96 kHz (the next is at 100 kHz).

 

[JLM1948]

I understand, but I thought the stepped sine measurement allowed much more resolution, like 1/96th octave. What am I confusing?
How can I produce a graph that extends up to 96k, or close enough?

 

[John Mulcahy]

I'll add 96k as an end frequency option in the next build.

 

[JLM1948]

Very nice, thank you!

 

[Antonio Di Motta]

I would like to point out the use of DSP HiFiBerry board to make room acoustic corrections.

Implementing Room acoustics correction using REW | HiFiBerry

 

[Antonio Di Motta]

Here is finally version 2 of the sound card, I have recently finished assembling and testing it and I can tell you that it was really a feat since the components used are all SMD in particular the passives are 0805 format. Fortunately, at a first test everything seems to work regularly, the power supplies are all present and stable.
Soon I will be sending tests and trials with REW.

Hello everybody
Antonio from Rome

 

[Shonver]

Congratulations on achieving this milestone, Antonio. I am following your project with great interest.

 

[Antonio Di Motta]

Thanks, I'm happy that my project could be interesting.
Now I attach some measurements that I made in loop output mode - input with sampling level from 48Khz up to 192Khz.
In this version of the board I used as a clock generator for the audio chip (CS4272) in 24.576 Mhz quartz frequency deviation equal to 20ppm, when I have the precision TCXO with 0.1ppm I will carry out the tests again.
Attached are the noise levels with sampling frequency respectively 48Khz - 96Khz and 192Khz.

 

[Antonio Di Motta]

As you can see the best level of performance is reached with the sampling frequencies at 48Khz and 96Khz while at 192Khz the card is much noisier above 40Khz. This behavior is most likely due to the cs4272 chip's DAC converter.
Now I attach some frequency response measurements at 48Khz and 96Khz sampling frequencies with output level at -3db from the full scale equal to about 0.7Vrms and also the compensated responses at the same sampling frequencies.

 

[Antonio Di Motta]

These are the distortion graphs measured with REW always the loop between input and output with different sample rates.
I hope I have not bored you, I still have many measures to post but for now I think it is enough.
Let me know what you think about your professional sound cards too.
Thanks for your attention