I think that I mentioned previously (Post 238) that I found that the two earlier tests were not conducted at the same output level from the DAC - and therefore into the ADC - for both software packages. REW displays the level differently than Multitone does. When I measured the output of the DAC using the Fluke 87, the output with REW was several dB higher than with Multitone. When I adjusted the two levels to be the same (.710 Vrms), the distortion results were pretty close, as shown in that post. Total Distortion plus Noise was within about a dB. I get that much variation in results when running the same test with the two software packs 10 minutes apart.
The readout from Multitone says that the crest factor for this test signal is about 5.83 (~15.3 dB). That comes out to be about 4.14 Vpk for a Vrms of .710 V. The DAC has a nominal full scale output of 4 Vrms for a sine wave - that's about 5.66 Vpk. So, when I ran the DAC level at .710 Vrms, that was a little less than 3 dB below clipping. That makes sense? When I was originally trying REW with the same waveform, it was much closer to clipping, and perhaps a bit beyond. That's why I decided to measure the RMS levels with both packages and make sure that they were the same for subsequent tests.
Yeah, the distortion products you highlighted are definitely higher with REW. Wouldn't you expect that when the DAC was run much closer to, and maybe even a little beyond, clipping? So, ignore that plot if you like. But, even when the levels were matched, REW shows some low level spikes that aren't visible in the Multitone plot. There could be a number of reasons for this, but I don't really know the exact cause.
The first examples weren't really meant to be definitive, but rather were just examples of what could be done. I should have made that clearer or just not posted them at all. Lesson learned.
The readout from Multitone says that the crest factor for this test signal is about 5.83 (~15.3 dB). That comes out to be about 4.14 Vpk for a Vrms of .710 V. The DAC has a nominal full scale output of 4 Vrms for a sine wave - that's about 5.66 Vpk. So, when I ran the DAC level at .710 Vrms, that was a little less than 3 dB below clipping. That makes sense? When I was originally trying REW with the same waveform, it was much closer to clipping, and perhaps a bit beyond. That's why I decided to measure the RMS levels with both packages and make sure that they were the same for subsequent tests.
Yeah, the distortion products you highlighted are definitely higher with REW. Wouldn't you expect that when the DAC was run much closer to, and maybe even a little beyond, clipping? So, ignore that plot if you like. But, even when the levels were matched, REW shows some low level spikes that aren't visible in the Multitone plot. There could be a number of reasons for this, but I don't really know the exact cause.
The first examples weren't really meant to be definitive, but rather were just examples of what could be done. I should have made that clearer or just not posted them at all. Lesson learned.
Since you mention it, CG, I believe your REW 17-tone measurement had some clipping or limiting. The signal level of -13.68 dBFS and the 'peak sample' value of -8.93 dBFS are not a factor 17 (12.3 dB, since power from the sines should be added, not amplitude, I believe) apart.
In order to eliminate all but one variable in the comparison process - the software spectrum analyzers - I used my own suggestion. I generated in Multitone an NID 17 tone sequence (as found natively in REW) and exported it to a wav file. I then played it back through my usual test DAC using Audacity. I used the Fluke 87 to confirm that output of the DAC was .710 Vrms. REW and Multitone were only used as spectrum analyzers. Same settings for FFT and so on.
Here is what the spectrum looked like in REW:
Now, here it is in Multitone:
As a sanity check and for completeness, I did the same for Jon Risch's Phi12r test. I did move the lowest tone down from 100 Hz to 97 Hz so that it wouldn't be at a subharmonic of the sample rate. Multitone measured the crest factor of the composite signal to be 4.9. The Fluke 87 said that the output level was .848 Vrms. That gave almost the identical peak output level as found in the NID test.
Here is the REW spectral plot:
Now, this is from Multitone:
I hope that this clears up any and all confusion and most questions.
Here is what the spectrum looked like in REW:
Now, here it is in Multitone:
As a sanity check and for completeness, I did the same for Jon Risch's Phi12r test. I did move the lowest tone down from 100 Hz to 97 Hz so that it wouldn't be at a subharmonic of the sample rate. Multitone measured the crest factor of the composite signal to be 4.9. The Fluke 87 said that the output level was .848 Vrms. That gave almost the identical peak output level as found in the NID test.
Here is the REW spectral plot:
Now, this is from Multitone:
I hope that this clears up any and all confusion and most questions.
For what it's worth (probably not much), the first sample shows indeed 12.3 dB difference between the (individual peak) signal and the 'peak sample' sum indicated in the top right REW box, but the second sample is already a little bit into clipping or saturation in REW, showing only a 10.8 dB difference.
Well, I don't know what to tell you.
The indicated peak sample levels were only 0.01 dB apart in level. I would think that if one is clipping, the other waveform would be clipping as well.
The two test sequences, NID 17 tone and Psi12r, have different crest factors as you might expect given the different structures. So, wouldn't you expect different RMS levels for the respective waveforms if the peak values are the same?
The indicated peak sample levels were only 0.01 dB apart in level. I would think that if one is clipping, the other waveform would be clipping as well.
The two test sequences, NID 17 tone and Psi12r, have different crest factors as you might expect given the different structures. So, wouldn't you expect different RMS levels for the respective waveforms if the peak values are the same?
Also, the indicated RMS value isn't necessarily what you might think > John Mulcahy on RMS
I'll quote from there:
"Please refer to the View preferences and the Full scale sine rms is 0 dBFS option: Full scale sine rms is 0 dBFS determines whether the rms level of a full scale sine wave is show as 0 dBFS or -3.01 dBFS. If this option is selected the rms level may be greater than 0 dBFS for some signal types, for example, a full scale square wave would have an rms level of +3.01 dBFS.
"This option is selected by default per the Audio Engineering Society conventions."
So, the actual RMS value is -3.01 dB lower than what is displayed, as I didn't change from the AES default setting. When you consider that, the crest factors for the two different test sequences come out almost exactly as Multitone reports.
In order to avoid confusion, I should probably change from the default for any plots I post. That might confuse some people who expect the default to be used. Or, I should just not post any plots.
I'll quote from there:
"Please refer to the View preferences and the Full scale sine rms is 0 dBFS option: Full scale sine rms is 0 dBFS determines whether the rms level of a full scale sine wave is show as 0 dBFS or -3.01 dBFS. If this option is selected the rms level may be greater than 0 dBFS for some signal types, for example, a full scale square wave would have an rms level of +3.01 dBFS.
"This option is selected by default per the Audio Engineering Society conventions."
So, the actual RMS value is -3.01 dB lower than what is displayed, as I didn't change from the AES default setting. When you consider that, the crest factors for the two different test sequences come out almost exactly as Multitone reports.
In order to avoid confusion, I should probably change from the default for any plots I post. That might confuse some people who expect the default to be used. Or, I should just not post any plots.
So true.
Basically, buy Panasonic industrial (not for retail sale) alkaline or Panasonic dry heavy duty. They are all metal encased and leak-proof (as far as leakproof can be). Do not ever buy alkaline Duracells or Energizers. As for Varta- DO NOT buy.
For everyone's benefit: You can get Panasonic Industrial at Mouser and Digikey. There's a Duracell Industrial as well.
I can't remember last I had a leaking battery. That's something I associate with batteries of the early 80s, before alkaline batteries became mainstream. That said, I replace the battery in my DMM so infrequently that getting a good battery that does not leak would make a lot of sense.
Tom
I can't remember last I had a leaking battery. That's something I associate with batteries of the early 80s, before alkaline batteries became mainstream. That said, I replace the battery in my DMM so infrequently that getting a good battery that does not leak would make a lot of sense.
Tom
OK, one last observation about these software packages.
I determined that the analyzers in Multitone and REW perform essentially identically, when set up identically. That's good news. No need to display any plots or numbers - they're the same. See post 283.
But, I did learn that the way Multitone and REW generate multitone test signals is different. I can't explain the details, because they're opaque to me.
The way I tested this was to generate that same NID pattern in REW and export it to a wav file. NID is a native test sequence built into REW. I'd previously reverse "engineered" (too strong a word - I just carefully determined the frequencies of the tones) the REW generated NID pattern and entered it all into Multitone. I then exported a wav file of the same ostensible test signal from Multitone.
Make sense so far?
Then I played both back through REW's RTA and Scope facilities using Audacity as the source. Since I'd already determined that the two application's spectrum analyzer functions performed the same, I didn't feel any need to replicate the tests using Multitone's analyzers. The hardware was the same as always: Topping E50 balanced output to an E1DA Cosmos Scaler to the E1DA Cosmos ADCiso.
First, the scope patterns. Multitone generated NID17 first, then REW Generated NID17:
There's immediately visible differences.
Now, this is the REW RTA spectrum analysis of the two. Again, Multitone generated NID17 first followed by REW generated NID17:
There's some difference in low level IM distortions visible in the two. But, although the RMS signal level is identical for both, the peak sample levels are very different. Why the distortion levels appear higher with a test signal with lower peaks beats the heck out of me.
I won't draw any conclusions other than to suggest that it's best to be consistent with what test signals you use across varying devices under test! OK, that and I think I've been able to reliably minimize any AC mains derived hum from being introduced into my test hardware.
My work here is done. (Contemporary meaning of the phrase.)
I determined that the analyzers in Multitone and REW perform essentially identically, when set up identically. That's good news. No need to display any plots or numbers - they're the same. See post 283.
But, I did learn that the way Multitone and REW generate multitone test signals is different. I can't explain the details, because they're opaque to me.
The way I tested this was to generate that same NID pattern in REW and export it to a wav file. NID is a native test sequence built into REW. I'd previously reverse "engineered" (too strong a word - I just carefully determined the frequencies of the tones) the REW generated NID pattern and entered it all into Multitone. I then exported a wav file of the same ostensible test signal from Multitone.
Make sense so far?
Then I played both back through REW's RTA and Scope facilities using Audacity as the source. Since I'd already determined that the two application's spectrum analyzer functions performed the same, I didn't feel any need to replicate the tests using Multitone's analyzers. The hardware was the same as always: Topping E50 balanced output to an E1DA Cosmos Scaler to the E1DA Cosmos ADCiso.
First, the scope patterns. Multitone generated NID17 first, then REW Generated NID17:
There's immediately visible differences.
Now, this is the REW RTA spectrum analysis of the two. Again, Multitone generated NID17 first followed by REW generated NID17:
There's some difference in low level IM distortions visible in the two. But, although the RMS signal level is identical for both, the peak sample levels are very different. Why the distortion levels appear higher with a test signal with lower peaks beats the heck out of me.
I won't draw any conclusions other than to suggest that it's best to be consistent with what test signals you use across varying devices under test! OK, that and I think I've been able to reliably minimize any AC mains derived hum from being introduced into my test hardware.
My work here is done. (Contemporary meaning of the phrase.)
Interesting that the higher crest factor generated less IM. Is there a way to bypass the DA-AD and feed the digital file into REW? My first suspicion is a dither issue or something similar in the creation side. I'll try to duplicate these but not for a few weeks when I'll be free to do this stuff.
I really appreciate all your work here. I know how tedious this type of testing and analysis can be.
I really appreciate all your work here. I know how tedious this type of testing and analysis can be.
Another possibility- how much out of band ultrasonic energy is generated with the different versions. Something I'll look at as well (using an analog spectrum analyzer).
REW can read and analyze a wav file directly.
https://www.roomeqwizard.com/help/help_en-GB/html/spectrum.html
RTA window, processing wav files.
It also shows phase of harmonic distortion components (@Markw4 )
https://www.roomeqwizard.com/help/help_en-GB/html/spectrum.html
RTA window, processing wav files.
It also shows phase of harmonic distortion components (@Markw4 )
Last edited:
If the signals were adjusted for an RMS level, that's different from adjusting for not clipping at the highest peak. Phase can affect RMS level and crest factor in somewhat different ways.
Last edited:
As suggested by PMA, here are the two generated test files measured at the wav level in REW - no hardware.
Now, as analyzed by Multitone:
It's too bad we actually need hardware to listen...
In a practical sense, I'm not sure what this all means. I guess just that Multitone and REW generate multitone test sequences differently, so you can't directly compare one to the other.
Now, as analyzed by Multitone:
It's too bad we actually need hardware to listen...
In a practical sense, I'm not sure what this all means. I guess just that Multitone and REW generate multitone test sequences differently, so you can't directly compare one to the other.
You might also analyze those wav files using a LUFS meter, which should use a reconstructed waveform to find peak intersample levels. As is, REW is showing rather different peak sample levels.
Please disregard my earlier comments about clipping. I made an unfounded assumption about the superposition of test tones and how that is reflected in the difference of 'peak sample' and the overall RMS amplitude.
CG, it occurs to me that REW might round off the display of the FFT-adjusted frequency values, i.e. chop off some digits. That could lead to not perfectly FFT-aligned test tone frequencies set for Multitone, which would then in turn lead to spectral loss, which could appear as increased intermodulation. Just a thought.
CG, it occurs to me that REW might round off the display of the FFT-adjusted frequency values, i.e. chop off some digits. That could lead to not perfectly FFT-aligned test tone frequencies set for Multitone, which would then in turn lead to spectral loss, which could appear as increased intermodulation. Just a thought.