Beta Test: Multitone Loopback Analyzer software

[Blumlein 88]

Would it be relatively simple in Multitone to have it show not just FR, but also phase? I know you do this in Deltawave which works quite well, but testing phase with a spaced tones using a test signal would be a nice addition. Or should I just keep using Deltawave for this?

 

[pkane]

Would it be relatively simple in Multitone to have it show not just FR, but also phase? I know you do this in Deltawave which works quite well, but testing phase with a spaced tones using a test signal would be a nice addition. Or should I just keep using Deltawave for this?

I’ve tried to get a reasonable approximation of phase response a while back from Multitone, but timing imprecision was messing it up to the point where it became too noisy and unusable. I suggest using DeltaWave for now. I’ll revisit phase with Multitone a bit later.

 

[MC_RME]

A few suggestions:

- Add option No window. Would be useful whenever the signal is synced and coherent. Also will then allow to use the trick that I explained elsewhere how to make noise-free filter response curves.

- Frequency response: I find it highly cumbersome when doing the measurement that the program always switches away from the Frequency tab and also rescales the plot. See I do FR after FR to see what changes in FR, and having to go back and rescaling just doesn't make sense to me.

- Is there a help or explanation what the Gain setting is doing?

- The two checkmarks to show blue and white spectrum, beside the Playback/Record selection/setup, are not inuitive to me. They should be separated and clearly labelled.

- Why is is that when the software starts it 'blinks' several times (and takes several seconds)? Before the program window finally comes up?

That said I find Multitone extremely useful. Comparing it with the APx it gives basically the same results, but adds the pure IMD/THD part that gives further detailed insight into the measurements. And its price / performance ratio is phantastic, obviously

 

[MC_RME]

Since the screenshot starts at 16kHz, if the noise generator can be set to have a 16kHz highpass, the plot can have a lower noise floor which can reveal the filters' shape more clearly I suppose? As well as 1000 tones from 16-22.05kHz with optimized crest factor...

I am pretty sure that doesn't work. The out-of-band display is imaging. Limiting the source signal would cause gaps in the area higher than sf/2.

 

[pkane]

A few suggestions:

- Add option No window. Would be useful whenever the signal is synced and coherent. Also will then allow to use the trick that I explained elsewhere how to make noise-free filter response curves.

- Frequency response: I find it highly cumbersome when doing the measurement that the program always switches away from the Frequency tab and also rescales the plot. See I do FR after FR to see what changes in FR, and having to go back and rescaling just doesn't make sense to me.

- Is there a help or explanation what the Gain setting is doing?

- The two checkmarks to show blue and white spectrum, beside the Playback/Record selection/setup, are not inuitive to me. They should be separated and clearly labelled.

- Why is is that when the software starts it 'blinks' several times (and takes severla seconds)? Before the program window finally comes up?

Thanks for the feedback, Matthias! There are a few things to try:

1. Option "No Window" is the same as "Dirichlet (Rect)" window -- that's just a less known name of rectangular window. It's already there. This will not work well if the clocks are not synced between DAC and ADC, and especially if these are running at two different sampling rates. But you can try it with ADI-2 Pro using the SPDIF clock sync, as I believe that scales the clock to other sampling rates. Maybe you can tell us if that works

2. Yeah, I think that switch back to spectrum plot is unnecessary. Don't recall why I even put that in...

A trick that works across all the plots if you don't want them to be rescaled:

Click and drag the plot a tiny bit, or zoom-in or zoom-out. This sets a fixed position and scale for the plot that will stop it from auto-scaling when new data is displayed. This works on all plots in Multitone (and same in DeltaWave, by the way). If you want to go back to auto-scaling, just click Reset Axis (should Axes, by the way, just remembered I need to change that!)

3. The software doesn't blink for me, but I suspect this has to do with ASIO drivers sending many update notifications and Multitone trying to update the driver list after receiving each one. Try disabling driver auto-update option to see if that's the case. I believe @Blumlein 88 had a similar issue. I may need to dig deeper into why this works this way for some drivers.

That said I find Multitone extremely useful. Comparing it with the APx it gives basically the same results, but adds the pure IMD/THD part that gives further detailed insight into the measurements. And its price / performance ratio is phantastic, obviously

That's great to hear! But due to inflation, the price is going up 100x in the near future

 

[MC_RME]

Thanks for the feedback, Matthias! There are a few things to try:

1. Option "No Window" is the same as "Dirichlet (Rect)" window -- that's just a less known name of rectangular window. It's already there. This will not work well if the clocks are not synced between DAC and ADC, and especially if these are running at two different sampling rates. But you can try it with ADI-2 Pro using the SPDIF clock sync, as I believe that scales the clock to other sampling rates. Maybe you can tell us if that works

3. The software doesn't blink for me, but I suspect this has to do with ASIO drivers sending many update notifications and Multitone trying to update the driver list after receiving each one. Try disabling driver auto-update option to see if that's the case. I believe @Blumlein 88 had a similar issue. I may need to dig deeper into why this works this way for some drivers.

View attachment 221200

Please put Rectangle at the top of the list. It equals 'no window' and as such shouldn't be in the middle of other windows listed. I just tried it - works. But for the filter response it won't. FFT size starts at 65k, that is much too much for such purposes and will cause accuracy issues. My test signal was 8k only, and is analyzed then at quad speed with 32k. Can you add those smaller numbers?

Disabling Detect driver changes did not change the startup behaviour. On my system it currently takes 16 seconds to start. I made a slow-mo video of the dialogs that blink up (to be able to read them). It turned out that Multitone calls up the 'Generic Low Latency ASIO driver', which then goes through all available audio devices checking them. This is a Steinberg tool coming with Cubase, you should ignore that one.

 

[MC_RME]

That's great to hear! But due to inflation, the price is going up 100x in the near future

Happy to pay!

 

[Blumlein 88]

Please put Rectangle at the top of the list. It equals 'no window' and as such shouldn't be in the middle of other windows listed. I just tried it - works. But for the filter response it won't. FFT size starts at 65k, that is much too much for such purposes and will cause accuracy issues. My test signal was 8k only, and is analyzed then at quad speed with 32k. Can you add those smaller numbers?

Disabling Detect driver changes did not change the startup behaviour. On my system it currently takes 16 seconds to start. I made a slow-mo video of the dialogs that blink up (to be able to read them). It turned out that Multitone calls up the 'Generic Low Latency ASIO driver', which then goes through all available audio devices checking them. This is a Steinberg tool coming with Cubase, you should ignore that one.

I haven't mentioned it because I thought it might be my 10 year old computer. It takes about 18 seconds to start up here. I unchecked the "check new versions", but that didn't make any difference. I don't remember when it first started being this slow, but has been slow the last 3 versions I think.

 

[bennetng]

I am pretty sure that doesn't work. The out-of-band display is imaging. Limiting the source signal would cause gaps in the area higher than sf/2.

If the filter is of the "slow" types yes, but almost every DAC chip has some sharper filters so the gaps will not be visible. Gaps are only visible with slow filters in analog tests, unless both the DAC and ADC have ridiculously low noise with almost no noise shaping. Current tests with full band white noise almost cannot show the difference between ESS linear phase fast roll off filter (120dB attenuation) vs Brickwall (100dB), especially when using typical audio interfaces for measurements instead of specialized devices like AP or Cosmos ADC.

 

[KSTR]

If the filter is of the "slow" types yes, but almost every DAC chip has some sharper filters so the gaps will not be visible. Gaps are only visible with slow filters in analog tests, unless both the DAC and ADC have ridiculously low noise with almost no noise shaping. Current tests with full band white noise almost cannot show the difference between ESS linear phase fast roll off filter (120dB attenuation) vs Brickwall (100dB), especially when using typical audio interfaces for measurements instead of specialized devices like AP or Cosmos ADC.

Full band periodic white noise isnt't optimal, that's true. But there are two relatively simple ways to get increased resolution with the same analyzer settings (FFT size etc).

1. Use a periodic full band signal with minimised crest factor. This the Noise Sweep with Newman Phases which ususally gives about -3dBFS RMS for typical FFT sizes.
It's an "endless" sweep where only one frequency is playing at a time except for the region at the stiching point where two frequencies are present.
8192 points, time domain:

Spectrum is flat (note I used 16k FFT size to better see the computation noise floor):

All bins below Nyqist are populated.

2. If we can use offline recording we may use time-domain averaging (assuming synced recording), the analyzer SW may even allow for that directly (without stopping the signal, of course).
If we are willing to wait until 100 (or even 1000) averages have completed, -10dB (or even -20dB) reduction of analog noise (and non-correlated digital noise) can be had.

----------

Further, reducing the bandwith of the Newman Sweep or other periodic noise (blocks of unpopulated bins) increases energy per frequency but we now have unpopulated regions in the images as well. Sometimes this is what we want (eg looking at potential artifacts in those blank regions if random noise can be kept low enough), sometimes not.

 

[Sokel]

I haven't mentioned it because I thought it might be my 10 year old computer. It takes about 18 seconds to start up here. I unchecked the "check new versions", but that didn't make any difference. I don't remember when it first started being this slow, but has been slow the last 3 versions I think.

It normally takes 4 seconds to start after pressing it but not always sometimes takes a lot longer.
Maybe has to do with the fact that I have 7 different devices but I always deactivate the ones I don't want to use leaving only 2 or three and I have never activated the Gigabyte multichannel realteck audio (through BIOS).
If that helps...

 

[pkane]

I haven't mentioned it because I thought it might be my 10 year old computer. It takes about 18 seconds to start up here. I unchecked the "check new versions", but that didn't make any difference. I don't remember when it first started being this slow, but has been slow the last 3 versions I think.

I suspect it has to do with testing of various ASIO drivers to check if device is connected or not. May be one or two old drivers that take a while or maybe just a lot of drivers.

 

[pkane]

Full band periodic white noise isnt't optimal, that's true. But there are two relatively simple ways to get increased resolution with the same analyzer settings (FFT size etc).

1. Use a periodic full band signal with minimised crest factor. This the Noise Sweep with Newman Phases which ususally gives about -3dBFS RMS for typical FFT sizes.
It's an "endless" sweep where only one frequency is playing at a time except for the region at the stiching point where two frequencies are present.
8192 points, time domain:
View attachment 221238

Spectrum is flat (note I used 16k FFT size to better see the computation noise floor):
View attachment 221240

All bins below Nyqist are populated.
View attachment 221241


2. If we can use offline recording we may use time-domain averaging (assuming synced recording), the analyzer SW may even allow for that directly (without stopping the signal, of course).
If we are willing to wait until 100 (or even 1000) averages have completed, -10dB (or even -20dB) reduction of analog noise (and non-correlated digital noise) can be had.

----------

Further, reducing the bandwith of the Newman Sweep or other periodic noise (blocks of unpopulated bins) increases energy per frequency but we now have unpopulated regions in the images as well. Sometimes this is what we want (eg looking at potential artifacts in those blank regions if random noise can be kept low enough), sometimes not.

Something like this, maybe?

 

[bennetng]

Beautiful!

 

[pkane]

Please put Rectangle at the top of the list. It equals 'no window' and as such shouldn't be in the middle of other windows listed. I just tried it - works. But for the filter response it won't. FFT size starts at 65k, that is much too much for such purposes and will cause accuracy issues. My test signal was 8k only, and is analyzed then at quad speed with 32k. Can you add those smaller numbers?

You can enter any value you want into the FFT-size setting on the main window. In fact, any of the values on this screen can be changed to something other than what's in the drop-down list (notice the 1024 averages setting, below).

Here's 4096 with white noise in 1.0.29

 

[pkane]

- Is there a help or explanation what the Gain setting is doing?

Just remembered that I didn't answer this.

Gain settings scales the output test signal by that many dB. You can also enter a fraction, negative or positive into the gain field.

 

[Sokel]

Maybe seems irrelevant,but I'm about to built a new PC and it would be nice to know what to choose so it has no problem in terms of "difficult" measurements.
Nothing specific,just that amount of memory,that CPU,etc.
To me is just for fun but still.
Thanks in advance!

 

[pkane]

Maybe seems irrelevant,but I'm about to built a new PC and it would be nice to know what to choose so it has no problem in terms of "difficult" measurements.
Nothing specific,just that amount of memory,that CPU,etc.
To me is just for fun but still.
Thanks in advance!

Memory - at least 16GB, more is better, faster is even ‘more better’

CPU (assuming Intel), pick the one with more GFLOPs performance:

https://www.intel.com/content/dam/s.../processors/APP-for-Intel-Core-Processors.pdf

More actual physical cores is better, but at least 4.

 

[Sokel]

Memory - at least 16GB, more is better, faster is even ‘more better’

CPU (assuming Intel), pick the one with more GFLOPs performance:

https://www.intel.com/content/dam/s.../processors/APP-for-Intel-Core-Processors.pdf

More actual physical cores is better, but at least 4.

Bookmarked this.
Thanks!

 

[Rantapossu]

Should the measurements in the measurement history be avaraged? They look very jagged for me...

And if you tick the different measurements on and off, everything is fine if you don't close the history selection window. But if you close the measurement history window and open it again, it doesn't remember the previous selections and you can't tick them off anymore. You'll have to close the history windows, clear all the graphs from the screen, open the history again and tick the measurements you want to show. When you do this, you will loose your latest averaged measurement and you can show only the jagged version of it.