I'm working on a setup to measure amplifier noise output, SNR, crosstalk etc. and see that a number of people are using a USB soundcard as the basis of such a setup.
Can anyone comment on improvements to the following setup?
Can anyone comment on improvements to the following setup?
- Scarlett 2i2 USB interface (this seems to be a popular USB interface). Are there USB interfaces with a better noise floor than this one (at a similar price level)?
- Is anyone using the TinySA?
- Oscilloscope 10:1 probe with 1 MOhm input impedance to connect between amplifier output and the USB interface. Will this impact the measured noise?
- Software - SpectraPLUS-RT seems good but I'd like a cheaper or free alternative, preferably that works cross platform.
A scope probe expects a 1M input impedance which the soundcard probably does not have (much lower) so you will be measuring wrong amplitude.
Using a 1:10 probe has the additional disadvantage that trhe large series R internally generates noise.
The best is a low impedance attenuator between amp output and sound card like 1k to 100 ohms. Low noise.
I'm using a Scarlet 2i2 which allows distortion and noise measurements down to about -100dB on a good day.
More than enough for tube stuff but not good enough for many ss units.
REW (don't be mislead by the name), ARTA, TrueRTA, is used by many here and is either free or low cost.
Jan
Using a 1:10 probe has the additional disadvantage that trhe large series R internally generates noise.
The best is a low impedance attenuator between amp output and sound card like 1k to 100 ohms. Low noise.
I'm using a Scarlet 2i2 which allows distortion and noise measurements down to about -100dB on a good day.
More than enough for tube stuff but not good enough for many ss units.
REW (don't be mislead by the name), ARTA, TrueRTA, is used by many here and is either free or low cost.
Jan
Most USB audio devices have a so called "high impedance" or "high Z" setting for guitars.
Most of the time this is 1M ohm.
But check your manual to be sure.
The Scarlet does have high-Z inputs.
Edit, according the website those are 1.5Mohm. So technically you could place a additional resistor parallel to make it fit 1Mohm.
Last edited:
ie Keep impedances low so scope type probes aren't needed. If some one wanted to analyse through each section of an amplifier a true scope probe would be needed.
The weakest area is likely to be the signal out of the interface. A tip from one particular software producer is that a true RMS DVM is essential. An example is setting the signal level. It can be used to set that while the source is loaded. A problem is that they generally cover a restricted freq range. Max of 500Hz isn't unusual so calibrate at that or lower. Some scopes can give a pretty accurate RMS reading, Depending on the level that can come out it may be possible to use an attenuator here as well.
Interface. I ran my Xfi usb sound card through this software input connected to output. This is the report on an earlier card
https://audio.rightmark.org/downloa...ing Methodology And Results For RMAA v5.5.pdf
The same sort of thing can be done on any interface. 🙁 A true sound card needs a bit of investigation. The spec's on a Scarlet may provide all of the info needed.
What level of SNR are you envisaging being needed to measure? What level of signal is being generated at the DUT output?
When you use a DVM for true RMS, then exact 1Meg matching for a scope probe isn't an issue per se, as the concern is more about the frequency response/flatness, and whether you can calibrate that out, and apply bandwidth filtering to suit you measurement scenario.
When you use a DVM for true RMS, then exact 1Meg matching for a scope probe isn't an issue per se, as the concern is more about the frequency response/flatness, and whether you can calibrate that out, and apply bandwidth filtering to suit you measurement scenario.
That's a toy, not likely to be useful for taking good measurements. For instance the only mention I can see of an FFT window function only mentions the Hanning window which is not used for measuring. You need to select a flattop window for accurate measurement of signal peaks.
Also it doesn't seem to mention noise measurements, which using an FFT have to be done differently to the spectral peak measurements as both the window function, sampling rate and number of points in the FFT affect the noise floor seen in a plot. For noise measurement the requirements of an FFT window are different.
I wrote some analysis tools in Python using the scipy.signal library, so for instance I save a .WAV from audacity, then post process for various things, often using FFT but not exclusively, for instance measure SiNaD or just distortion. Something I can do (not actually using the FFT it happens) is determine the test tone freq/amp/phase by refining a model iteratively, then subtract it from the data to plot the residual distortion+noise, which can be very revealing.
So when looking for some software be on the lookout for a tool that knows about spectral peak measurement using flattop window(s), and separate noise measurements, and ideally distortion-residual plotting.
Useful reading about measuring with an FFT: https://holometer.fnal.gov/GH_FFT.pdf
https://www.roomeqwizard.com/help/help_en-GB/html/spectrum.html
That one uses the java runtime. A pretty sensible option. There are other more appropriately named general analyser packages around written for windows but will run on emulators such as wine on linux however free versions of ones I have noticed are almost bound to get people reaching for their wallet for upgrades.
RightMark will run under wine but no longer supported. Just available but not it's pro version. The free one will generate the loop back sound card report I posted.
Some moons ago I came across web sites with details on using java for FFT
Last edited:
I use Visual Analyser https://www.sillanumsoft.org/ mainly for FFT, and Audio Tester https://www.audiotester.de/ for frequency response measurement with EMU 0404 USB soundcard.
If you want to measure higher voltages than 1-3V RMS, must to use divider or another protection device.
I use AC millivoltmeters with 1V output or pmillet's "Sound Card Interface".
If you want to measure higher voltages than 1-3V RMS, must to use divider or another protection device.
I use AC millivoltmeters with 1V output or pmillet's "Sound Card Interface".
A bit late to the party, but I'm developing a usb audio analyser (https://beembedded.com/behydra/), so let me know if you still need something and happy to have you use one of the beta samples in case it helps!
João, as you can read above everyone seems to be interested in getting to the last dB of whatever is possible.
Could you post some FFT's and specifications for our community? Any pricing information?
Could you post some FFT's and specifications for our community? Any pricing information?
Have a look here: https://neurochrome.com/pages/measuring-distortion-on-the-cheap
While I focus mainly on distortion, the same considerations apply for the measurements you mention. Of the various softwares available, I found REW to be powerful and relatively straight-forward to use.
Tom
While I focus mainly on distortion, the same considerations apply for the measurements you mention. Of the various softwares available, I found REW to be powerful and relatively straight-forward to use.
Tom
I looked at your web page and its interesting but light on implementation details. Can you share more?
At the moment I’ve not implemented with 32-bit USB, so the audio FFT’s will look unimpressive (and there still some unwanted noise I need to remove on the anaolgue side) with -140dBFS unweighted non-integrated. I’ll promise to share some FFT’s when they’re likely to be more impressive 🙂
At the moment, due to the beta users I have, my focus has been more on the digital and wireless protocols, which means the analogue audio will be more towards the end (but feedback on what you normally need and prioritise for testing/development would be greatly appreciated!). The path to take on the analogue audio side is what I'm sturggling a bit to come up with the best solution/compromise. Is your development/testing entirely on the analogue audio only or do you also tend to use digital protocols?
On the product price, when the time comes, I’ll advertise the product on the marketplace on diyAudio, so I don’t want to focus too much on the product advertising so I don’t break any forum rules. But just for reference, price-wise, it will vary on features so that everyone can get what they need at reasonable prices, aiming between £450 to £950 (here would include all features, Bluetooth BR/EDR, LE Audio, I2S/PCM/TDM, SPDIF, Analogue Audio, PDM). Essentially the Bluetooth features will push the price to the £950 because of qualification costs.
Thanks, and of course, happy to share and answer where I can, even though from a hardware perspective there are not many details yet to share. Essentially on the implementation, I’m creating a modular hardware approach (as well as software naturally) in order to allow for easy hardware upgrades in case needed, reduced costs and e-waste for myself and the user. This is the main concept of the product.
In terms of the digital wired/wireless protocols, not really much to say as the solution is very software-centric. At the moment I have the protocols I2S/PCM (not yet TDM), PDM (input only), Bluetooth A2DP/HFP/AVRCP, LE Audio (some of it, not everything is yet available), SPDIF and then Analogue audio in/out. Here essentially the software is configured so that the user can configure via a GUI or CLI each protocol as input and/or output and with the configurations supported as per specification.
On the Analogue In/Out, is at the moment where I’ll be struggling the most I think (naturally). At the moment for development speed purposes, I’m using the built-in ADC & DAC from the SoC, with an auxiliary external LNA (which the goal is to give a ~-165dBV noise floor, unweighted un-integrated) in case the user wants to record the noise floor of their setup, but at the moment this would come at the expense of clipping if the input to the LNA would already be at 1Vrms.
The audio analogue side it’s where I’ve been struggling the most to decide what should be my approach and therefore, I’m quite(extremely) curious to understand people’s views on it. What I’m considering at the moment is having 2 modules: one where the audio analogue in/out is more limited and with a -140dB noise floor and another one where there’s a bit extra hardware to improve the performance.
As there already are affordable audio analyzers (Quant Asylum QA403 or Cosmos devices) and even open source devices suitable for audio measurements (https://www.diyaudio.com/community/threads/diy-adcs.419922/post-7844591) what would be the main selling point of your analyzer?
I had already seen your work in https://www.diyaudio.com/community/threads/diy-adcs.419922/post-7844591) and it's amazing!
At the moment the main advantages of it are the digital wired and wireless protocols, which you can use as an audio analyser or if you want it can also be configured as a data generator.
For enterprise purposes, I would say that the automation will also be an advantage, since it allows an engineer to design the test system, for example, for audio algorithms and a second engineer to focus on the development of the algorithm itself.
Then ideally so that the user can have everything packed into one, the audio analogue will also need to be good,
Apart from APx, on the digital side (wired & wireless), I've not quite found something that can do something similar without having to hook-up a development board (ESP, STM or even Qualcomm) to configure and use. That said, I may be unaware of something currently that covers the digital side, but please do let me know in case I've missed something :/
At the moment the main advantages of it are the digital wired and wireless protocols, which you can use as an audio analyser or if you want it can also be configured as a data generator.
For enterprise purposes, I would say that the automation will also be an advantage, since it allows an engineer to design the test system, for example, for audio algorithms and a second engineer to focus on the development of the algorithm itself.
Then ideally so that the user can have everything packed into one, the audio analogue will also need to be good,
Apart from APx, on the digital side (wired & wireless), I've not quite found something that can do something similar without having to hook-up a development board (ESP, STM or even Qualcomm) to configure and use. That said, I may be unaware of something currently that covers the digital side, but please do let me know in case I've missed something :/