Problems using MI

[Testman]

Problems using MI

Hi,

I set up a system using MI and a Juli@ card. Win7 32bits, 4GB, Athlon II X2.

When I was testing loopback mode in order to see the residuals of the system on THD and THDA, I found out an unstable behaviour on the results, showing sometimes good values as expected for the card and many times poor values on THD, SNR etc.

I did again the same tests using other (free) analyser software and the results are good, consistent and always the same using the same adjust levels and configs as I did on MI.

For example, the values I achieved using MI, using 24bits, 48kHz, ASIO mode were 0,0007% THD and 99.8dB SNR for a few moments during the tests. Most of the times I achieved 0,0017% THD and 77dB SNR, very perceptible on spectrum analyser.

The best value I had using MI I had too all the times using the other software and MME drivers. I tried MME drivers on MI, with no better results..

Is there any place in this forum where I can post the screenshots of MI results and configurations?

Best regards,
MarceloYared
Post at 8-9-2011 04:02

Simultaneous sound card input & output consumes more CPU time and USB bandwidth (if a USB sound card is used), especially when the bit depth and sampling rate are high. Here are some tips if you encounter unstable problems:

1. To reduce the CPU time consumption, in Multi-Instrument, you can set the Trigger Mode to “Auto” (i.e. free run for the case of a sound card) if triggering is not required. THD measurement does not require triggering.

2. Use integer test frequency in the Signal Generator if possible to reduce the CPU time consumption.

For THD measurement, there are two options regarding the test frequency:

Option 1: use the so-called "No Spectral Leakage" test frequency, together with the Rectangle Window function in the Spectrum Analzyer. This option provides the best results if the oscilloscope (i.e. sound card input) and the signal generator (i.e. sound card output) share the same sampling clock, which is true if the same sound card is used for both input and output. Otherwise option 2 should be used.

Option 2: use any test frequency, preferably an integer test frequency (in order to reduce the CPU time consumption of the signal generator). When a test frequency with spectral leakage is used, Kaiser 6 Window function is recommended in the Spectrum Analyzer for THD measurement. Rectangle Window function should not be used in this case.

If you do not know what the “no spectral leakage” frequency is, just enter the test frequency you want in the Signal Generator and then tick the “no spectral leakage” option. The Signal Generator will then calculate the “no spectral leakage” frequency for you based on the current sampling frequency and FFT size. In most of cases, the “no spectral leakage” frequency is not an integer value. With a “no spectral leakage” test frequency, Rectangle window function should be used in the Spectrum Analyzer. Otherwise, Kaiser 6 window function is recommended. For both cases, the record length of the Oscilloscope should be set to a value equal to or greater than the FFT size to avoid zero padding.

3. When ASIO driver is used, you can go to [Setting]>[Display]> to set “ASIO Buffer Size” to “Max” if your encounter unstable problem.

4. If a USB sound card is used, choose a USB2.0 port of the computer to connect to the sound card.

By the way, you can upload an image as an attachment in this forum.

virtins
Post at 8-9-2011 14:35

Tks for your help!

Now I have excellent and stable results at 48kHz! Better than the results using the freeware.

I´m working on 96kHz and 192kHz, using the same configurations, but without good results until now.

In fact Juli@ is a PCI card, and the CPU usage is fine (no more than 50%). The problem was in the configurations I did.

And, if it can help, I found out in your faq the instructions to upload images, but I can´t understand it. I think a little bit more explanations there would help.

Regards and tks again,
MarceloYared
Post at 10-9-2011 23:16

Looks like you have resolved the issue with 96kHz and 192kHz.

virtins
Post at 13-9-2011 20:52

Hi,

Its me again.

Actually I solved the problems using the Kaiser 6 window and ftest an integer.

But I´m not satisfied with the bad behaviour using Rectangle window and no spectral leakage or non integer frequency as appears to be a default for MI.

Because this I can´t use the DTP feature, since it is using Rectangle window.

My system has good performance, latency constant and better than 500us and the performance monitor show no lack of resources to run MI, as you can see in the pictures in annex.

Is there anything I can do to solve it?

Best regards,
MarceloYared
Post at 8-10-2012 07:23

[Testman]

Hi,

Any news?

If it is bug of MI, at least can you show me how to change the window type in DTP plans in order to allow the use of the feature?

Best regards,

MarceloYared
Post at 13-10-2012 23:44

It is not a bug. You can save a Panel Setting File with the THD measurement settings you preferred. In DTP, use instruction "LDP" (Load a Panel Setting File) to load this Panel Setting File before you start the THD measurement. Please refer to DTP sample files such as "AudioParameter_SR48000_A.dtp" for reference.

virtins
Post at 28-10-2012 09:16

Ok, Tks again,

But I still have no answer to the bad performance using rectangle window and/or non integer frequencies.

Regards
MarceloYared
Post at 9-11-2012 06:51

The cause of "bad performance using rectangle window and non integer frequencies" in your particular setup could be:

1. This option requires that the signal generator (DAC) and the oscilloscope (ADC) hardware share the same sampling clock, which is generally not a problem for a sound card. But whether this is the case for your sound card or not is unknown.

2. It is time consuming for the computer to generate a non-integer frequency signal. If the computer is not up to speed (CPU or USB communication speed), some discontinuities may be observed in the captured waveform in the oscilloscope. You can check whether this is the case. If this happens, you can lower the sampling rate, use single channel, lower the sampling bit resolution, use USB 2.0 port, adjust the ASIO buffer size if ASIO driver is used...

virtins
Post at 11-11-2012 18:08

Tks again,

Actualy I´m using an Asus Xonar STX and Juli@, both PCI cards, with Asio drivers
The behaviour is consistent using Kaiser 6 and integers for frequencies (always good performance) as is consistent (always bad performance) when using rectangle and/or non integers, no matter how I change the parameters om MI.
Can you show me the figure above with better resolution in order to see the configurations you´re using toi achieve those results? I´ll try to reproduce it here.
And, can you tell me what is the PC configuration you´re using too?
Best regards,

MarceloYared
Post at 11-11-2012 21:23

Please refer to the following test reports for details:
http://www.virtins.com/doc/D1004/EMU_Tr ... _D1004.pdf
http://www.virtins.com/doc/D1007/EMU_0204_Report.pdf

virtins
Post at 12-11-2012 22:07

Ok, tks
I see the tests were done using external sound cards with USB interfaces, so, the conditions are quite diferent from mine.

If you don´t mind I will rephrase my question in this way:
- Is there any configuration using internal PCI or PCIx audio cards that works using MI in rectangle window and/or non integer frequencies configurations? If yes, can you show me where I could find the specifications of the system and setup of MI in order to reproduce them here?

best regards and thanks in advance,
MarceloYared
Post at 13-11-2012 02:26

We do not have a report with a PCI sound card yet. But in terms of communication speed, PCI should be better than USB. Did you zoom in and check the captured waveform in your oscilloscope windows and see whether you have :
(1) discontinuity in the captured sine waveform. If yes, it means a CPU/communication speed problem.
(2) If [FFT size] = 32768, then check whether an exact integer number of cycles are contained from POINT 1 to POINT 32768 using the cursor reader. If not, it means the ADC part of the sound card does not share the same sampling clock with the DAC part of the sound card and there are some disparity between the two hardware sampling clocks. In this case, a Kaiser 6 window has to be used.

virtins
Post at 13-11-2012 11:53

Ok, Tks,

Actually I did this using a multiple frequency in order to see the correct beginning and end of the waveform on scope.

Using rectangle window I have the first picture below.

If I understand well your instructions, there are no discontinuities (if not, actually does not matter the window type, it would happen using Kaiser 6 too) and the scope shows 33 complete cycles of a 480.3Hz sinus (actually 480Hz in the generator panel). The problem is still there.

In the second I have the same frequency but using Kaiser 6 and 480Hz. the results are consistent with the 16bits depth and are ok.

It seems to me that the sound card is OK. I did the same procedure with a Juli@ sound card. Same results.

It seems to me too like a problem with the software, since I have excellent results using RMAA (Hanning window) at any bit depth.

If I´m not doing everything correct I need to know, because the system itself is working fine, but I can´t use all the features of MI, a very good software IMHO.

Regards and thanks in advance,

MarceloYared
Post at 13-11-2012 23:31

[Testman]

Thank you for the info. Now, it is clearer. This is not a software bug. The software works perfectly indeed.

In your Fig. 1, the signal frequency measured in the spectrum analyzer is 4800.3Hz (according to you, the signal frequency in the signal generator panel is 4800Hz), but in the oscilloscope window, "33 complete cycles" of the signal is displayed in a 1- second long frame. Obviously, this is an "Alias" effect in display. If the signal frequency is 4800Hz, then there should be 4800 cycles displayed in 1-second long frame. If the signal frequency is 480Hz, then there should be 480 cycles displayed in the 1-second long frame. If you only count from POINT 1 to POINT 32768 (i.e. a 32768-point FFT segement), then there should be 32768/48000* 4800= 3276.8 cycles for the case of 4800Hz and 32768/48000* 480= 327.68 cycles for the case of 480Hz. If you (1) zoom sufficiently or (2) right click anywhere within the oscilloscope window, select "oscilloscope chart option" and choose "display all data points" , the "alias" effect will disappear.

Note that in both cases (4800Hz or 480Hz), the FFT segement (32768 points) contains a non-integer number of cycles, which requires you to use Kaiser 6 to suppress the spectral leakage when you want to measure the harmonic distortion. Again, in MI, for THD measurement, it is suggested that a rectangle window function can be used only when an integer number of cycles are contained in a FFT segment (not the oscilloscope frame), otherwise Kaiser 6 must be used instead. Hope the following three figures can help to make the above idea clearer. The three figures were obtained by digital loopback at software level (ideal case), which means THD should be very much close to zero if the measurement parameters are set properly.

(1) Fig. 1, signal frequency =4800.29296875Hz, the FFT segement (32768 points) contains 32768/48000*4800.29296875 =3277 cycles, rectangle window function can be used.

(2) Fig. 2, signal frequency =4800Hz, the FFT segement (32768 points) contains 32768/48000*4800 =3276.8 cycles, rectangle window function is used with THD measurement accuracy affected quite significantly.

(3) Fig. 3, signal frequency =4800Hz, the FFT segement (32768 points) contains 32768/48000*4800 =3276.8 cycles, Kaiser 6 window function is used to suppress the spectral leakage in order to measurement THD accurately.

As fas as we know, RMAA uses a non-rectangle window function to suppress the spectral leakage regardless of whether an integer number of cycles are contained in the FFT segment or not. Therefore it is always on the safe side. But it may introduce a little distortion when an integer number of cycles are actually contained in the FFT segement. On the other hand, MI gives the user much more options and flexibility in handling different situations.

virtins
Post at 15-11-2012 13:13

[Testman]

Ok, tks again,

Actually the frequency I used was 480Hz, but it is not a problem here.

In the real life the measurements without the sound card only shows that the software is OK, except in the IO to DAC and from ADC.

I tested the system using the default parameters (rectangle and non integer freqs) and for 16bits and one channel I achieved satisfactories results. So, I tried 24bits and two channel configurations. After many tweaks I achieved satisfactories results with a minor misbehaviour with Xonar.
I had to:

- uninstall all features that controls the clock and bus of the Mobo in order to have energy saving (Asus softwares, in my case);
- put the windows in the performance mode at energy mode configurations at control panel);
- change the latency of the Xonar to the maximum (80ms);
- change the latency of the Juli@ to the maximum (2048 samples);
- put the trigger in AUTO mode and restart the spectrum analyser after started the signal generator.

As you know Xonar Essence and Juli@ are top class sound cards, and seems to me Virtins could spend a little more time in order to increase the performance using internal sound cards, maybe working on the synchronism or buffering.

At the end I´m trying to modify the parameters of your DTP archives to use 24bits and change the windowing and signal levels, but with no success until now. Could you elaborate a bit more your explanations above for me?

Tks in advance and best regards,

MarceloYared
Post at 17-11-2012 02:22

As explained in the second post, if you encounter any unstable issue, try to put the trigger mode to "Auto" and ASIO buffer size to Max (via [Setting]>[Display]>[ASIO buffer size]), lower the bit depth, reduce the number of channels, decrease the sampling rate, use integer frequency in the signal generator (therefore Kaiser 6 window function has be to used), in order to lower demand for computing power and communication speed.

virtins
Post at 17-11-2012 17:45

But actually there is no sense in to stress the performance of the system in order to obtain a marginal stability using MI in the default mode.

I tried the suggestions in your second post. They failed without unninstalling the monitoring softwares specially on 24bits depth.

And the performance monitors show the system running in low demand, in both modes, in IO, CPU usage and memory usage. The latency is always constant and low.

Again, I think you have to strongly consider to increase the performance of the software using PCI or PCIx communication, because, as you say, both are much faster than USB. It seems to me there is something that can be improved in the syncronism with the ADC/DAC using internal devices.

And btw I learned how to modify/create the DTP profiles and I would suggest two things:
- to create the capacity of plotting two channels in the same chart and;
- to edit only one of the parameters in each line when one is changing them (the editor replaces all fields at the same time).

Best regards,
MarceloYared
Post at 17-11-2012 19:35

Thanks for the suggestion. The second suggestion has already been implemented in Verssion 3.3 which will be released soon.

virtins
Post at 20-11-2012 08:33

Hello and happy new year!

For 5 years i used the PCI-x Asus Xonar Essence STX on a powerfull PC running in XP x64 OS. It should be noted that i've also tried a top external sound card the EMU-0404 USB, but the performance of Asus Xonar was much better. Four months ago i did an upgrade of my PC: OS=Win7 Pro 64bit, CPU=Intel i3 2120 3.3GHz, RAM=DDR3 8GB. From the description you could understand that this system not only fulfils but easily surpass the demands of any modern software. I've been also confirmed for this by the two SPICE simulators that i use.
Unfortunatelly i have simillar issues like those described from Marcelo Yared. In the new system, with the ASUS drivers upgraded as well its ASIO driver, i can do THD measurements only using Kaiser 6 window. I can't use rectangle window. And the difference is signifficant in real device measurements. With the previous PC the THD of a single ended tube amplifier measured from MI 3.2 using rectangle window was 1.235% and now with the new PC using Kaiser 6 window is 1.624% (1Wrms/8Ohms). Likewise the MI measurement of bandwidth of this amplifier was 40KHz(-3dB) and now is just.... 22KHz. The performance of Asus Xonar in No Loopback test, is also worst than in previous system.
I am sure that is an issue of the ASIO driver of sound card and how is handled by MI. Yesterday i changed the drivers of Asus Xonar with the old used in XP x64 and the MI indicated the old good performance (THD=0.0004%, NL=-117dBFs, bandwidth=98KHz etc) in No Loopback test. But the software is unstable, it stops and closed by Win7. Please note that is not a system crash! With the new drivers and ASIO of sound card installed, in No Loopback test with Kaiser 6 smoothing window MI shows a THD=0.0015%, NL=-96dBFs and bandwidth=50 to 60KHz. I've tried any possible combination of Asus drivers and MI 3.2 configurations but nothing.

Best regards
Fotis Anagnostou
Post at 4-1-2013 17:39

First of all. Happy New Year!

We have not tested MI with this ASUS sound card. It may be too big to fit into the casing of our desktop PC. Did you try the MME driver instead of the ASIO driver?

virtins
Post at 6-1-2013 00:58

Thank you so much for the wishes!
I could not believe that you don't have available a modern desktop PC!
Regarding MME drivers are useless for the field of our measurements. Yes, are working for some time but.... the results. We use MI for Hi-End Audio measurements and so only a top performance sound card - like the Asus Xonar STX Essence - could be the right interface to MI. Because MI is the best software - FFT analyzer as it allways gives the most trustable measurement results! Please take a look in two different threads and maybe you could understand our specific problems:

http://www.diyaudio.com/forums/analogue ... otios.html

http://www.diyaudio.com/forums/analog-l ... s-fun.html

Thank you so much!

Fotis
Post at 10-1-2013 20:01

ASUS drivers for Win7 are rubbish

Hi to all
After a last exhaustive research, i could confirm you that the ASUS drivers - i.e. the C-media Oxygen ASIO driver - for Win7 are wrong. There are 4 different driver upgrades offered by ASUS and i've tried one by one without any improvement in FFT tests. The function "No spectral leakage" can't be used. When is enabled, in oscilloscope window i can see a constantly repeated bounce per second, it seems like oscilloscope is re-triggered continuously, is unstable and similarly the FFT analyzer. There is problem of synchronization. The only solution is the use of smoothing window i.e. "Kaiser 6".
Another one issue is the bad frequency response. In Bandwidth Test the upper -3dB frequency limit lies at 50 to 60KHz!
All above tests are executed in "No Loopback" mode at 192Ks / 24bit with ASIO = auto.
I am completely disapointed, in my previous PC with "WinXP x64" OS everything were OK! Now, with "Win7 64bit Pro" OS the ASUS XONAR Essence STX is useless, its performance in "No Loopback" tests is unacceptable. I can't use MI Pro 3.3 anymore. The only solution is to spend another 150 to 200 Euro to buy a new sound card and to throw the ASUS XONAR in the trash bin.
I am so sorry.

Fotis
Post at 13-7-2013 14:35

"No Spectral Leakage" option requires more communication bandwidth between the software and the sound card as the signal frequency is normally not an integer, maybe the ASUS's new driver does something wrong when handling higher communication traffic. Anyway, Kaiser 6 option should be good enough for THD measurements.

virtins
Post at 15-7-2013 19:36

just got a screenshot from the Editor-in-Chief, http://www.canardpc.com/, he used ASUS XONAR Essence STX.

virtins
Post at 3-10-2013 12:44

[khushfehmi045]

In the real life the measurements without the sound card only shows that the software is OK, except in the IO to DAC and from ADC.
I tested the system using the default parameters (rectangle and non integer freqs) and for 16bits and one channel I achieved satisfactories results. So, I tried 24bits and two channel configurations. After many tweaks I achieved satisfactories results with a minor misbehaviour with Xonar.
I had to:
- uninstall all features that controls the clock and bus of the Mobo in order to have energy saving (Asus softwares, in my case);
- put the windows in the performance mode at energy mode configurations at control panel);
- change the latency of the Xonar to the maximum (80ms);
- change the latency of the Juli@ to the maximum (2048 samples);
- put the trigger in AUTO mode and restart the spectrum analyser after started the signal generator.
As you know Xonar Essence and Juli@ are top class sound cards, and seems to me Virtins could spend a little more time in order to increase the performance using internal sound cards, maybe working on the synchronism or buffering.??

[VirtinsTech]

If you use ASIO driver and encounter some "discontinuity" in the captured waveform, go to [Setting]>[Display] and set "ASIO Buffer Size" to MAX. Also, Change the Trigger Mode to "AUTO". For ASIO driver, the Device Model in both [Setting]>[ADC Device] and [Setting]>[DAC Device] should be set to "Sound Card ASIO".