Conrad,
the noise was my first question!
How come EMU specifies 112dB os signal to ratio?
It is barely impossible to make measurements on a phono stage as the gain at 1KHz it is usually over 50dB unless i filter the input then the noise floor starts to go down...but still....i don't like it.
i think i am going to build the Cordel signal generator for this purpose.
anyways...i was thinking of the same...meaning......i don't know...maybe with external power supply or some settings....i might get the output quite....but i wouldn't be so sure.
Anyways......amazing.....the inut can go as down as 0.0003%....excellent!
the noise was my first question!
How come EMU specifies 112dB os signal to ratio?
It is barely impossible to make measurements on a phono stage as the gain at 1KHz it is usually over 50dB unless i filter the input then the noise floor starts to go down...but still....i don't like it.
i think i am going to build the Cordel signal generator for this purpose.
anyways...i was thinking of the same...meaning......i don't know...maybe with external power supply or some settings....i might get the output quite....but i wouldn't be so sure.
Anyways......amazing.....the inut can go as down as 0.0003%....excellent!
fotios, I don't have that software, but can assure you via traditional measurements, the THD+N answer is simply wrong. If it's any consolation, the THD+N number in Visual Analyzer appears to be completely wrong as well. I think they may have misplaced a decimal by three places- 67% instead of 0.067%, but I don't have much confidence in that either.
On the USB ports, I was too embarrassed to mention it, but I had all kinds of trouble until I realized that the front panel USB connections of my Acer are probably 1.1 and the rear panel is 2.0. One should be suspicious of any USB ports that are associated with a camera/memory card reader. The E-MU is a disaster of non-responding drivers and software with USB 1.1, and only a full reboot will fix it.
I'm still perplexed about the output noise. Nothing I've done with software or hardware level controls has had any effect at all. Loading doesn't affect it. It just is, yet I can't believe E-MU could accept something so bad- they generally make very good stuff.
On the USB ports, I was too embarrassed to mention it, but I had all kinds of trouble until I realized that the front panel USB connections of my Acer are probably 1.1 and the rear panel is 2.0. One should be suspicious of any USB ports that are associated with a camera/memory card reader. The E-MU is a disaster of non-responding drivers and software with USB 1.1, and only a full reboot will fix it.
I'm still perplexed about the output noise. Nothing I've done with software or hardware level controls has had any effect at all. Loading doesn't affect it. It just is, yet I can't believe E-MU could accept something so bad- they generally make very good stuff.
Conrad Hoffman said:
Then, why should one pay 1250 USD in PHS for an unreliable software like SpectraPlus? As for Visual Analyzer, at least is offered free, and great thanks deserves to its designer for his offer.
As for the output noise issue, thanks to you, i did a measurement of my EMU0404 output noise so you can make an estimation, a comparison or whatever you want. The oscillograph has taken with the output volume fully CCW (-oo). If the volume increased then the curves worsen. The CH1 curve has taken in unbalanced mode via a mono jack inserted in output A, and the CH2 curve has taken in balanced mode via a stereo jack inserted in output B but from its tip terminal only.
Anyway, many thanks Conrad for your kindness to discuss with me about these oppresive issues.
Regs
Fotios
An externally hosted image should be here but it was not working when we last tested it.
Just some random Sunday morning thoughts- No matter the cost of the software, the THD+N number just has to be significantly larger than the THD number. If they're almost equal, something is wrong, or you have the quietest DUT ever made. I have to wonder if there's some as yet undiscovered bug in the E-MU driver. It's almost like there's some extra noisy source being mixed in that I can't turn off. The basic output simply can't be this bad.
There was a question way back about windowing the FFT data. I don't know if VA correctly implemented their windows, but they seem to work correctly. Based on what I see with those, you'd want a Hanning (not Hamming) or maybe a Blackman for these measurements, not a flat top or any of the various other choices, lest the measurement results will be larger than they should be. Windowing is outside my expertise, though I've implemented several in software I've written, but a Google search of FFT Windowing should yield a wealth of info.
There was a question way back about windowing the FFT data. I don't know if VA correctly implemented their windows, but they seem to work correctly. Based on what I see with those, you'd want a Hanning (not Hamming) or maybe a Blackman for these measurements, not a flat top or any of the various other choices, lest the measurement results will be larger than they should be. Windowing is outside my expertise, though I've implemented several in software I've written, but a Google search of FFT Windowing should yield a wealth of info.
Conrad Hoffman said:
Hi Conrad
In the manual of V.I. i found a very interesting guide regarding smoothing windows. I attach the text as is:
"The above parameters are automatically computed based on the peak frequency value detected in each frame of data. In order to achieve high measurement accuracy, the frequency of the test signal, the sampling frequency of the Oscilloscope, the Record Length and the FFT size must be carefully chosen such that a FFT segment contains exactly an integer number of cycles of the test signal, in order to avoid any artificial noise introduced due to the spectral leakage inherent in FFT algorithm. It is recommended to use the following formula to derive the frequency of the test signal: N × [Sampling Frequency]/[FFT Size], where N is an integer. For example, when the Sampling Frequency is 44100 Hz and the FFT size is 16384, and you want a test signal of about 1 kHz, then the recommended test frequency would be 1001.293945 Hz (i.e. N=372) and the Record Length should be set to a value greater than the FFT size to avoid zero padding. The following table lists the recommended test frequencies for different sampling frequencies and FFT sizes. No window function is necessary if the recommended test frequency is used."
Regs
Fotios
That is interesting! It does, however, emphasize the care one has to use when making these sorts of measurements- the devil is in the details (and just about everywhere else he can hide).
I want to make an A-weighted measurement of the E-MU output. I can do it with a series of filtered measurements and a spreadsheet but that's a PITA. Does anybody have a schematic for a proper A-weighting filter I could build to just put in front of a wide-band meter?
I want to make an A-weighted measurement of the E-MU output. I can do it with a series of filtered measurements and a spreadsheet but that's a PITA. Does anybody have a schematic for a proper A-weighting filter I could build to just put in front of a wide-band meter?
I believe that we don't have any reason to prompt our measurement stuff with false arguments. The true is true and the false is false. We must know what we have precisely in our hands.
For last time, i made a final measurement of E-MU0404 with M.I. in the suggested settings from its manual, to make a comparisson with the measurement of output noise with my DSO. I quote the two pictures to you make your calculations
Some usefull conclusions:
1. The EMU control panel shows that is internally locked at 192KHz, but the measurement done in 48KHz. This proves that the meas software commands the EMU device and not the control panel of EMU for the sampling rate. Maybe this is an idioma of M.I. because it is compatible with the ASIO drivers of the ADC/DAC devices of EMU. I repeat that when MI is launched the first thing that does is the searching for ASIO drivers. You can see in the top line of its window that the ASIO is loaded: < ASIO EMU 0404 I USB >
2. The record length of the sample, is callibrated automatically from MI according to sampling rate chosen. You can see that is 10 times bigger than FFT size chosen.
3. The rest settings are: a) Weighting=Flat, to can be comparable with the DSO measurement b) Average=Linear of 20 frames, because either 10 or 200 frames averaged, the resulted levels are same. There is not reason for long time waiting to completed a 100 or 200 samples averaging. To be sure i placed 20.
4. The output level of EMU placed at its minimum possible because the DSO meas done with output level at -oo (pot fully closed). You can see in the top of MI window the level at 2% for precise results.
5. The window used it is Flat Top for precise measurement of noise level. Moreover, if you have read my previous post, when the frequency is placed at a precise value (1000,488821 Hz) then there is not need for smoothing window.
6. The two horizontal dot lines presents continuously the power level of noise level exists during FFT proccesing regardless of the rest operations executed during.
If we make a calculation of the better obtained noise level from DSO which is of CH1 in unbalanced mode (it is better than balanced of CH2!!! and this is not curious because the ICs and parts used inside EMU for inverting the signal to get the cold terminal) we resulted in a noise level of: a) The small component which is a normal output noise is about: 3mVpp = 1.06mVrms = -59dBV b) The large oscillations which occurs every 250ìs are: 15mVpp = 5,3mVrms = -45,5dBV. For a complette period of 200ìs, the mean value is about -53dBV. The value shown in M.I. is NL = -61,22dBV. The same level show also the dot horizontal lines in the middle of window.
Well, i think the values obtained from either the DSO and the M.I. are very close between them. The difference of about 7dBV it is obviously random because the two measurements done seperately. Moreover 7dBV represents a different of 1mV.
Good-bye gentlemen
Fotios
For last time, i made a final measurement of E-MU0404 with M.I. in the suggested settings from its manual, to make a comparisson with the measurement of output noise with my DSO. I quote the two pictures to you make your calculations
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
Some usefull conclusions:
1. The EMU control panel shows that is internally locked at 192KHz, but the measurement done in 48KHz. This proves that the meas software commands the EMU device and not the control panel of EMU for the sampling rate. Maybe this is an idioma of M.I. because it is compatible with the ASIO drivers of the ADC/DAC devices of EMU. I repeat that when MI is launched the first thing that does is the searching for ASIO drivers. You can see in the top line of its window that the ASIO is loaded: < ASIO EMU 0404 I USB >
2. The record length of the sample, is callibrated automatically from MI according to sampling rate chosen. You can see that is 10 times bigger than FFT size chosen.
3. The rest settings are: a) Weighting=Flat, to can be comparable with the DSO measurement b) Average=Linear of 20 frames, because either 10 or 200 frames averaged, the resulted levels are same. There is not reason for long time waiting to completed a 100 or 200 samples averaging. To be sure i placed 20.
4. The output level of EMU placed at its minimum possible because the DSO meas done with output level at -oo (pot fully closed). You can see in the top of MI window the level at 2% for precise results.
5. The window used it is Flat Top for precise measurement of noise level. Moreover, if you have read my previous post, when the frequency is placed at a precise value (1000,488821 Hz) then there is not need for smoothing window.
6. The two horizontal dot lines presents continuously the power level of noise level exists during FFT proccesing regardless of the rest operations executed during.
If we make a calculation of the better obtained noise level from DSO which is of CH1 in unbalanced mode (it is better than balanced of CH2!!! and this is not curious because the ICs and parts used inside EMU for inverting the signal to get the cold terminal) we resulted in a noise level of: a) The small component which is a normal output noise is about: 3mVpp = 1.06mVrms = -59dBV b) The large oscillations which occurs every 250ìs are: 15mVpp = 5,3mVrms = -45,5dBV. For a complette period of 200ìs, the mean value is about -53dBV. The value shown in M.I. is NL = -61,22dBV. The same level show also the dot horizontal lines in the middle of window.
Well, i think the values obtained from either the DSO and the M.I. are very close between them. The difference of about 7dBV it is obviously random because the two measurements done seperately. Moreover 7dBV represents a different of 1mV.
Good-bye gentlemen
Fotios
fotios, I'm not seeing your pictures- hosting site bandwidth exceeded perhaps? Anyway, I've sent an email to E-MU for clarification on how to confirm their output noise specification. When the paperwork says -118dB A-weighted, I expect to be able to measure and confirm it with traditional analog instruments. FWIW, I'm not sure we're seeing the same controls with our software, as the two models are a bit different. We'll see if they reply!
OK, some more comments. First, on windows, and not the usoft one. To check and make sure you have read the peak amplitude of the fundamental OK, you should use a flattop window. All THD measurements are a ratio of the fundamental (i.e. say 1kHz) to the harmonics, and if you want accuracy in that measurement, you should at least use both and see if there is a difference. Notice I say THD and NOT THD + noise. Due to the nature of the hanning window, there can be an amplitude error of up to 16% if you happen to get it worst case.
How to tell - measure the say 1kHz peak using hanning, then switch to flattop. If the amplitude increases when going to flattop, you have just seen the hanning error.
Now, if you read where the hanning window should be used, it's usually in 2 cases. First, for best frequency resolution - i.e. close in sidebands, and second, for measuring random noise. So I was partially wrong when I said to use Flattop above. It should be used in all cases to measure THD, to make sure you measure the amplitudes of the fundamental and harmonics correctly. But for THD+noise, you probably want to use the hanning to measure the random noise correctly, and note the error in making the peak fundamental measurement (per switching windows above) and add it to the figure.
Yes, you can get tricky and make the input frequency an exactly perfect ratio of the sample frequency so that no window is required (uniform window). But I should point out this also requires reliable triggering on the waveform, and trigger delay such that the sine wave starts precisely at a 0 crossing for the first sample, and ends at a zero crossing for the last sample. This may be accomplished in sound cards automatically when the source DAC and input ADC are both running from the same sample clock and data acquisition start and stop are syncronized. Or not. But will not be the case for an external signal from more pure sources like your Cordell generator. Any slow drift of that external signal will also muck it up.
Noise - by definition, it is a function of bandwidth. As a signal to noise (SNR) measurement, it is meaningless unless the bandwidth over which the noise measurement is made is also stated. In many cases you may recall the inherent noise of a device is stated for a 1Hz bandwidth, "per Hz" or /Hz. Multiply times BW to get total noise.
In many cases the noise bandwidth is not stated, but it's implied or common to some industries like pro or consumer audio where IEEE standards for making noise definitions exist. The one that is used the most, that makes the equipment look the "best" is A weighted, which is supposed to correspond to the way humans process noise.
We are more sensitive mid and upper mid noise than at the extremes. So if you see audio noise measurements that look pretty good and no standard is given, assume it is A weighted. Marketing departments at work. To make any sense of the THD + N measurements in software, you need to know the formula or definition used to make the measurement. My first thought when I see THD+N better than THD is that A weighting is likely used, and it may be an error where A weighting is applied before both the THD and N is calculated.
As a reality check for THD, you can always measure the difference from the fundamental to the tallest harmonic. If the tallest harmonic is more than 6-10dB greater than the others, this single difference will be very close to the correct total THD except in cases of many, many harmonics extending out a long way.
Started working with FFT analyzers in 1981, and have been involved with them in a professional way until today. I have an HP3562A and Ono Sokki CF-6400 as my main workhorses, but also use soundcards with FFT software for different things. I've recently been playing around with importing data from sound cards to Agilent VEE, which gives limitless back end processing and display capabilities. But not for the faint of heart. I also use an HP 54522A scope with a great built in FFT capability. These have come way down in price as of late, I paid $600. USD for it. This is light years above USB scope solutions, as it has 2 channels of 2GS/s and 500 MHz BW. It's also set up for HP 1144A active probes, which come in handy if you ever do digital audio projects
How to tell - measure the say 1kHz peak using hanning, then switch to flattop. If the amplitude increases when going to flattop, you have just seen the hanning error.
Now, if you read where the hanning window should be used, it's usually in 2 cases. First, for best frequency resolution - i.e. close in sidebands, and second, for measuring random noise. So I was partially wrong when I said to use Flattop above. It should be used in all cases to measure THD, to make sure you measure the amplitudes of the fundamental and harmonics correctly. But for THD+noise, you probably want to use the hanning to measure the random noise correctly, and note the error in making the peak fundamental measurement (per switching windows above) and add it to the figure.
Yes, you can get tricky and make the input frequency an exactly perfect ratio of the sample frequency so that no window is required (uniform window). But I should point out this also requires reliable triggering on the waveform, and trigger delay such that the sine wave starts precisely at a 0 crossing for the first sample, and ends at a zero crossing for the last sample. This may be accomplished in sound cards automatically when the source DAC and input ADC are both running from the same sample clock and data acquisition start and stop are syncronized. Or not. But will not be the case for an external signal from more pure sources like your Cordell generator. Any slow drift of that external signal will also muck it up.
Noise - by definition, it is a function of bandwidth. As a signal to noise (SNR) measurement, it is meaningless unless the bandwidth over which the noise measurement is made is also stated. In many cases you may recall the inherent noise of a device is stated for a 1Hz bandwidth, "per Hz" or /Hz. Multiply times BW to get total noise.
In many cases the noise bandwidth is not stated, but it's implied or common to some industries like pro or consumer audio where IEEE standards for making noise definitions exist. The one that is used the most, that makes the equipment look the "best" is A weighted, which is supposed to correspond to the way humans process noise.
We are more sensitive mid and upper mid noise than at the extremes. So if you see audio noise measurements that look pretty good and no standard is given, assume it is A weighted. Marketing departments at work. To make any sense of the THD + N measurements in software, you need to know the formula or definition used to make the measurement. My first thought when I see THD+N better than THD is that A weighting is likely used, and it may be an error where A weighting is applied before both the THD and N is calculated.
As a reality check for THD, you can always measure the difference from the fundamental to the tallest harmonic. If the tallest harmonic is more than 6-10dB greater than the others, this single difference will be very close to the correct total THD except in cases of many, many harmonics extending out a long way.
Started working with FFT analyzers in 1981, and have been involved with them in a professional way until today. I have an HP3562A and Ono Sokki CF-6400 as my main workhorses, but also use soundcards with FFT software for different things. I've recently been playing around with importing data from sound cards to Agilent VEE, which gives limitless back end processing and display capabilities. But not for the faint of heart. I also use an HP 54522A scope with a great built in FFT capability. These have come way down in price as of late, I paid $600. USD for it. This is light years above USB scope solutions, as it has 2 channels of 2GS/s and 500 MHz BW. It's also set up for HP 1144A active probes, which come in handy if you ever do digital audio projects
That is not what i have been told by SpecraLab.
They told me flat window for frequency analysis and Hanning for THD....
interesting....
Anyways.....Has anybody ever tried to figure out why there is such a huge wide band noise at the output?
I started this 3d with this big question mark.....BTW i am really contend my 3d is being a usefull source of information for many people.......
unfortunately we still haven't understood how to get rid of the output noise...at least....
i just can't believe that none of you guys has ever tried to feed an high voltage gain amplifier with a since wave produced by the EMU and scope the output!
It is super noisy....of course....but how come in a loop test i can go down to -150dB???? this is a bit odd i guess.....
Anyways....i would like to build a very low noise/low THD signal generator to use along with the input of the soundacard that accordingly with Conrad's measurements (surely reliable) it can be precise and can measure very low THD.
If anybody has some suggestion on a simple signal gen to build....i will gladly consider it.
I was thinking about Cordell signal generator but the issue here is how to find for cheap all these steppoed attenuators.....mmmm.....that's why i would fold back on something that doesn't require thise things.
any tip is very welcome
They told me flat window for frequency analysis and Hanning for THD....
interesting....
Anyways.....Has anybody ever tried to figure out why there is such a huge wide band noise at the output?
I started this 3d with this big question mark.....BTW i am really contend my 3d is being a usefull source of information for many people.......
unfortunately we still haven't understood how to get rid of the output noise...at least....
i just can't believe that none of you guys has ever tried to feed an high voltage gain amplifier with a since wave produced by the EMU and scope the output!
It is super noisy....of course....but how come in a loop test i can go down to -150dB???? this is a bit odd i guess.....
Anyways....i would like to build a very low noise/low THD signal generator to use along with the input of the soundacard that accordingly with Conrad's measurements (surely reliable) it can be precise and can measure very low THD.
If anybody has some suggestion on a simple signal gen to build....i will gladly consider it.
I was thinking about Cordell signal generator but the issue here is how to find for cheap all these steppoed attenuators.....mmmm.....that's why i would fold back on something that doesn't require thise things.
any tip is very welcome
Just now i did a check of the output of EMU0404, but this time in direct monitor mode - i.e. stand alone, without connection to PC - and the DSO shows an output noise level from 2mVrms to 4,4mVrms. That means: from -53,98dBV min to -47,13dBV max. Also those groups of oscillations presented in the oscillograph of my previous post, does not exists... instead single peaks produced every 270ìsec (f=3,7KHz) with a magnitude of 14mVpp max. This has the sense that the USB cable and the PC connection does not affect the overal noise figure.
Fotios
Fotios
syn08 said:
Sorry but i don't get it.
In what sense would I see the 1sample noise?
If i connect the output at the scope with a 1KHz sine wave on 24bit 48KHz mode i do have the same noise doesn't go down whatsoever...and that is what feeds the input of my circuit...a noisy analogic sine wave.....what does the 1sample have to do with the sine wave at the output?
BFNY said:
Hello BFNY
Many thanks for your precious help and advices. It is obvious that you are the most expertised person from all of us in such type issues which discused in this thread.
If you remember, in one post i reffer that Virtins (please don't take it as advertisement, i fear maybe missunderstand me all of you in the end) recently presented a cheap and nice USB interface such as a DSO of 8bit at 100MSa/s. Because its price was only 170 euro, i asked the company for its compatibility as well with the FFT of MI in which included also a scope. The answeer that i got was, that it is better for FFT analysis the use of a proffesional sound card. The reason is the dynamic range according to formula: DR = 20*log (bit depth in decimal / 1). For example, an ADC with 8bit res = 1byte = FF hex = 256 Dec... according to the above formula a 8bit ADC has DR = 20*log(256/1) = 48dB. For a 16bit ADC accordingly the DR = 96 dB and for a 24bit ADC the DR = 144dB.
The sales conductor said me, that an 8bit independently if its sampling rate it is 100MSa/s or 2GSa/s, due to its limited dynamic range can't gives reliable results in THD, THD+Noise, SNR, IMD, NL measurements.
I have checked just now that your HP54522A has an ADC of 8 to 10bit resolution. Maybe is there some secret of your nice device?
Also i have checked the Pico Scope site, and they reffer the same as Virtins, that their better device for FFT analysis was the 216 of 16bit (800USD).
Really i am confused after your last post. What can i do? To by or not to by this ADC of Virtins?
Regards
Fotios
Stefanoo said:
You asked about the noise floor at -150dB, even if the input is noisy, and the answer is in the RBW concept.
Why is the D/A generated signal so noisy, that's a different story and nobody said it should or could be much better. Filter that signal (which is, in the analog space, the equivalent of going from 1 sample to "n" samples) with a reasonable BW (of say 20KHz) and see if you still dislike the result. At best, don't use it.
already filtered it...but still noise in on the way...that's why i was asking if anybody in case was aware of a low THD signal generator easy to build.syn08 said:
As i said cordell was an excellent candidate if it wasn't for all those stepped attenuator that are hard to find!
Thanks anyways for yor hint
You are seeing residual noise on the output because the analog signal is created (most likely) by a very high frequency one bit converter and then filtered. The output filter after the DAC used gets most, but not all the noise attenuated. This noise is high frequency, hence you can see it on your scope. Perhaps they economized on the output filter.
So why is is not a problem on the loopback? There are anti-alias filters on the inputs that attenuate the high frequency noise. In a nutshell, the filters on the inputs seem to be better/steeper, with more poles, than the ones on the output. This is best illustrated if you could bypass, or turn off the inputs analog filters. If you could, you would see all that noise. Even though it was at higher frequencies, it would be "aliased back" into the audio range.
There is also another possibility. Remember this is a digital consumer grade circuit. There could be coupling from high frequency clocks and other devices inside the box that is leaking to the output through the ground plane or power supply.
So why is is not a problem on the loopback? There are anti-alias filters on the inputs that attenuate the high frequency noise. In a nutshell, the filters on the inputs seem to be better/steeper, with more poles, than the ones on the output. This is best illustrated if you could bypass, or turn off the inputs analog filters. If you could, you would see all that noise. Even though it was at higher frequencies, it would be "aliased back" into the audio range.
There is also another possibility. Remember this is a digital consumer grade circuit. There could be coupling from high frequency clocks and other devices inside the box that is leaking to the output through the ground plane or power supply.
Stefanoo said:
- Linear Technology AN43 application note.
- Linear Technology AN67 application note.
- Tektronix SG505 schematic http://www.diyaudio.com/forums/showthread.php?postid=502295#post502295
- State variable oscillator http://www.uwm.edu/~msw/AudioOsc/index.html
If you don't decide to buy a second hand SG505, I would build the last one in the list.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.