Lavry Gold Savitr (new A/D standalone)

[pkane]

Yes, I was getting to the point that what you hear is what you hear is mostly intermodulation distortion which by the way one of the things AP uses to measure quickly, because they space the multitude logarithmicaly.

But my point is that the loop listening test does "hide" the harmonic distortion, making the test questionable to me. (One can measure intermodulation with dual tones).

My main problem with loop testing is that some people do use some free software to MATCH AD and DA converters, which is nonsense to me. I see it has been going for about 10 years. They seem to dismiss AP testing. I encountered claims that it does what the AP can't do.

I am retired comfortably, but it bothers me to see more than a few people believing such stuff. That free software is based on measuring, not listening. I think the software is broken. The "proof" that it works is done by people listening afterwords to the recomanded matched converters.

Dan, do you really want to continue the discussion from Gearspace here? There's nothing wrong with a well-measuring ADC being used to measure a DAC or other devices in a loopback. As long as the ADC is tested thoroughly (let's say it's been measured by an APx555) and shows very low distortion in the desired bandwidth, flat frequency response and non-existent jitter. Why would one not be able to use it to measure a DAC or another device? AP sells software just for this purpose (APx500).

 

[rdenney]

Yes, I was getting to the point that what you hear is what you hear is mostly intermodulation distortion which by the way one of the things AP uses to measure quickly, because they space the multitude logarithmicaly.

But my point is that the loop listening test does "hide" the harmonic distortion, making the test questionable to me. (One can measure intermodulation with dual tones).

My main problem with loop testing is that some people do use some free software to MATCH AD and DA converters, which is nonsense to me. I see it has been going for about 10 years. They seem to dismiss AP testing. I encountered claims that it does what the AP can't do.

I am retired comfortably, but it bothers me to see more than a few people believing such stuff. That free software is based on measuring, not listening. I think the software is broken. The "proof" that it works is done by people listening afterwords to the recomanded matched converters.

We are all arguing against counterproductive myth and lore, but sometimes we forget to be specific and explicit about the myth and lore we are arguing against.

Of course, if DACs and ADCs are already so linear and undistorted that there's nothing to correct by matching (beyond precisely and cleanly matching gain, which is where pro equipment provides features more useful than a lot consumer equipment).

This is a big philosophical issue in my line of work (about which I need to be a bit coy). My professional colleagues pretty frequently confuse objectives with performance measures, but in the absence of appropriately stated objectives, performance measures become the objectives. The objective should state what we are trying to attain in terms relevant to a non-technical user, and the performance measure should demonstrate (technically) that we have attained it. Confusing the two opens the door to measurements that have no real utility to users becoming the target feature.

For example, my objective with audio equipment is to make no audible alteration to the signal present in the recording. This objective is not shared by everybody, of course, but I think it is pretty broadly accepted by ASR regulars. We use measurements of features like noise and distortion to demonstrate to what extent the device under test attains that objective. I think this concept is consistent with the point you made that distortion is a secondary effect, and we should be concerned with the primary effect of how it sounds. I would restate your point that distortion is an output feature, while audibly clean sound entering our ears is an input requirement. That is good advice: confusing objectives and performance measures is confusing the gozintas with the gozouttas.

For those who do not state or understand the objectives clearly, a low THD number or whatever might become the objective, and that's the door through which products like the software you describe can enter into the thinking of an otherwise well-considered audio enthusiast. But it can get much worse than that. If the objective is simply to please me, the listener, then suddenly the biases baked into what pleases me--including my habitually favorite distortions if you will--become the objective and we are off the golden-ear races.

Rick "often in the middle of design arguments that should be objectives arguments" Denney

 

[Dan Lavry]

Dan, do you really want to continue the discussion from Gearspace here? There's nothing wrong with a well-measuring ADC being used to measure a DAC or other devices in a loopback. As long as the ADC is tested thoroughly (let's say it's been measured by an APx555) and shows very low distortion in the desired bandwidth, flat frequency response and non-existent jitter. Why would one not be able to use it to measure a DAC or another device? AP sells software just for this purpose (APx500).

This is a more scientifically oriented forum. I think many people here are capable of understanding that the concept of matching converters is flawed. And what I saw was NOT about using a good known AD to evaluate DAs. it was about MATCHING AD and DA.

In a loop, the AD does the conversion first. The converted music a arrives at the DA later, so it is too late for the DA to correct an AD distortion. The DA considers the all the incoming signal as data to convert; it can’t separate the distortion from the music.

But even if it could, it does not have any means of connection to the AD. The AD has no idea what the DA is about to do to the signal.

This is a more scientifically oriented forum. I think many people here are capable of understanding that the concept of matching converters is flawed. And what I saw was NOT about using a good known AD to evaluate DAs. it was about MATCHING AD and DA.

In a loop, the AD does the conversion first. The converted music a arrives at the DA later, so it is too late for the DA to correct an AD distortion. The DA considers the all the incoming signal as data to convert; it can’t separate the distortion from the music.

But even if it could, it does not have any means of connection to the AD. The AD has no idea what the DA is about to do to the signal.

So how do you match converters? One answer I got from you so far was a long the line of “I don’t understand so there is no point trying to explain to me”. That is not an answer.

I have issues with listening to loops; I explained that it is a poor test because it masks the harmonics. I agree that many loops will eventually show up as audible (mostly inter-modulations). But I certainly have a problem with using a test to MATCH converters.

What is missing is an EXPLANATION of how it works. Please explain how a loop test can match converters (independent unit). Connecting units in a loop can’t even separate a poor unit from a good one (measure or listen, the outcome is dictated by the poor unit, hides the good one).

I value the information from a properly set up listening test by qualified people. But I am more than reluctant to accept opinions for total strangers when it conflicts with my technical understanding and don’t line up with some top-notch mastering and recording engineers tell me.

Say the AD generates 2KHz harmonic distortion and the DA generates 3KHz. The outcome is 2KHz and 3KHz distortion. The DATA (music) fed back to does not change the converters behavior. So how does that matching work? It calls for an explanation.

The SHIFT to doing the test using a good AD converter to test DA is NEW and recent. It was not about using a good AD, it was about MATCHING AD and DA's. The software did not claim to just replace the AP generator as a source to evaluate DA's. It was about running loops in order to figure out AD and DA MATCHING. It recommends Ads, not just DAs. My impression is that is what many people still think it does. I did not previously see that requirement for a high-quality AD for that test. People go into a thread and they see years of “matched pair” recommendations, not just a couple of top notch AD driving various DA, but many ADs to match the DAs.

 

[antcollinet]

This is a more scientifically oriented forum. I think many people here are capable of understanding that the concept of matching converters is flawed. And what I saw was NOT about using a good known AD to evaluate DAs. it was about MATCHING AD and DA.

In a loop, the AD does the conversion first. The converted music a arrives at the DA later, so it is too late for the DA to correct an AD distortion. The DA considers the all the incoming signal as data to convert; it can’t separate the distortion from the music.

But even if it could, it does not have any means of connection to the AD. The AD has no idea what the DA is about to do to the signal.

This is a more scientifically oriented forum. I think many people here are capable of understanding that the concept of matching converters is flawed. And what I saw was NOT about using a good known AD to evaluate DAs. it was about MATCHING AD and DA.

In a loop, the AD does the conversion first. The converted music a arrives at the DA later, so it is too late for the DA to correct an AD distortion. The DA considers the all the incoming signal as data to convert; it can’t separate the distortion from the music.

But even if it could, it does not have any means of connection to the AD. The AD has no idea what the DA is about to do to the signal.

So how do you match converters? One answer I got from you so far was a long the line of “I don’t understand so there is no point trying to explain to me”. That is not an answer.

I have issues with listening to loops; I explained that it is a poor test because it masks the harmonics. I agree that many loops will eventually show up as audible (mostly inter-modulations). But I certainly have a problem with using a test to MATCH converters.

What is missing is an EXPLANATION of how it works. Please explain how a loop test can match converters (independent unit). Connecting units in a loop can’t even separate a poor unit from a good one (measure or listen, the outcome is dictated by the poor unit, hides the good one).

I value the information from a properly set up listening test by qualified people. But I am more than reluctant to accept opinions for total strangers when it conflicts with my technical understanding and don’t line up with some top-notch mastering and recording engineers tell me.

Say the AD generates 2KHz harmonic distortion and the DA generates 3KHz. The outcome is 2KHz and 3KHz distortion. The DATA (music) fed back to does not change the converters behavior. So how does that matching work? It calls for an explanation.

The SHIFT to doing the test using a good AD converter to test DA is NEW and recent. It was not about using a good AD, it was about MATCHING AD and DA's. The software did not claim to just replace the AP generator as a source to evaluate DA's. It was about running loops in order to figure out AD and DA MATCHING. It recommends Ads, not just DAs. My impression is that is what many people still think it does. I did not previously see that requirement for a high-quality AD for that test. People go into a thread and they see years of “matched pair” recommendations, not just a couple of top notch AD driving various DA, but many ADs to match the DAs.

I don't think anyone here - when talking about multiple loops of conversion have ever talked about matching converters.

I'm feeling your preaching to the choir a little.

 

[Dan Lavry]

I don't think anyone here - when talking about multiple loops of conversion have ever talked about matching converters.

I'm feeling your preaching to the choir a little.

Thank you for saying that. I hope you are correct. But I think there has been conversation about the test on this forum about evaluating AD/DA loops by means of Audio Diffmaker.

The concept of AD-DA loop left a “bad taste in my mouth”. I first noticed it (on another forum) when the test included my gear. I disagreed with the result. At first, I was angry, then I asked enough questions resulting in ZERO answers and much worse nonsense! People did not want to hear about it. I guess it is not pleasant to admit that you spend money based on some wrong information. So I stopped asking questions.

The software used in the DA-AD matching test uses digital data comparison. You start with a digital file feeding DA and send the result to AD. Digital to digital enables comparing files on a computer.

So here my first question (this does not require technical knowledge, it is a “user” question):

Does the output file contains data in the bandwidth all the way to Nyquist (48KHz for 96K sampling, 96KHz for 192K)?

If the answer is YES, the data covers all the bandwidth, then noise shaping gets “bad grade” instead of “good mark”. Other energy above audio (in most of the “unusable” converter bandwidth) indicates poor results. That is a bad test.

if the answer is NO, the software filters out above 20KHz. That calls for altering the data with a filter, a whole other question. But does testing 96KHz sampling, is done using a 20KHz filter? So why bother with 96KHz and 192KHz when this kind of testing requires limiting the bandwidth to 20KHz?

Audio converters focus on the hearing range say up to 20Khz or a bit higher. It is not uncommon to shift noise (noise shaping) to higher frequencies, it benefits the audible range but “contaminate” the frequency above (The Specification and FFT show it). That and other low-level activity at very high frequency is not much of a problem for audio.

I have more questions, but I would really like an explanation to that not very technical one, so most people can understand the question and the answer.

 

[pkane]

I think many people here are capable of understanding that the concept of matching converters is flawed

I'm still puzzled by this statement. This is what you were saying on the other forum, but why do you think that anyone is proposing any sort of DA/AD matching? I've yet to see anyone claim such a thing, so there's no disagrement.

Does the output file contains data in the bandwidth all the way to Nyquist (48KHz for 96K sampling, 96KHz for 192K)?

The file that's used in that Gearspace thread for all their testing is a recorded orchestral piece at 44.1k with most of the content below 20k. For generalized null test, you decide what sampling rate to use and what content for your specific testing needs.

if the answer is NO, the software filters out above 20KHz. That calls for altering the data with a filter, a whole other question. But does testing 96KHz sampling, is done using a 20KHz filter? So why bother with 96KHz and 192KHz when this kind of testing requires limiting the bandwidth to 20KHz?

I don't use the software that's used in that thread, but my own app doesn't filter either the reference or the recorded file. The differencing is done in the time domain. You can apply arbitrary filters if you want during the analysis, but there's no filtering by default.

 

[Newman]

We are all arguing against counterproductive myth and lore, but sometimes we forget to be specific and explicit about the myth and lore we are arguing against.

Of course, if DACs and ADCs are already so linear and undistorted that there's nothing to correct by matching (beyond precisely and cleanly matching gain, which is where pro equipment provides features more useful than a lot consumer equipment).

This is a big philosophical issue in my line of work (about which I need to be a bit coy). My professional colleagues pretty frequently confuse objectives with performance measures, but in the absence of appropriately stated objectives, performance measures become the objectives. The objective should state what we are trying to attain in terms relevant to a non-technical user, and the performance measure should demonstrate (technically) that we have attained it. Confusing the two opens the door to measurements that have no real utility to users becoming the target feature.

For example, my objective with audio equipment is to make no audible alteration to the signal present in the recording. This objective is not shared by everybody, of course, but I think it is pretty broadly accepted by ASR regulars. We use measurements of features like noise and distortion to demonstrate to what extent the device under test attains that objective. I think this concept is consistent with the point you made that distortion is a secondary effect, and we should be concerned with the primary effect of how it sounds. I would restate your point that distortion is an output feature, while audibly clean sound entering our ears is an input requirement. That is good advice: confusing objectives and performance measures is confusing the gozintas with the gozouttas.

For those who do not state or understand the objectives clearly, a low THD number or whatever might become the objective, and that's the door through which products like the software you describe can enter into the thinking of an otherwise well-considered audio enthusiast. But it can get much worse than that. If the objective is simply to please me, the listener, then suddenly the biases baked into what pleases me--including my habitually favorite distortions if you will--become the objective and we are off the golden-ear races.

Rick "often in the middle of design arguments that should be objectives arguments" Denney

Important post well stated. Minor pedantic quibbles on the philosophy paragraph, not for ASR.

Newman "strategic management tragic" Newman!!

 

[Dan Lavry]

I'm still puzzled by this statement. This is what you were saying on the other forum, but why do you think that anyone is proposing any sort of DA/AD matching? I've yet to see anyone claim such a thing, so there's no disagrement.


The file that's used in that Gearspace thread for all their testing is a recorded orchestral piece at 44.1k with most of the content below 20k. For generalized null test, you decide what sampling rate to use and what content for your specific testing needs.


I don't use the software that's used in that thread, but my own app doesn't filter either the reference or the recorded file. The differencing is done in the time domain. You can apply arbitrary filters if you want during the analysis, but there's no filtering by default.

Say some AD or DA takes advantage of the concept of noise shaping. It is a pretty old concept widely used. Say some noise shaping at 96KHz sample rate “moves” 5dB of the noise to 24KHz-48KHz frequency range. In fact, there is a slope with the level growing with increased frequency. That or any other high frequency energy is a problem.

If you run a loop 10 times, the noise will accumulate to 50dB. That will show up in the digital comparison as a huge problem. That noise is in the data, poor converter gets an F.

My second question: Does your software require to synchronize the data source to the DA clock. This is not a common AES or SPDIF DA, it will requires separate clock input to the DA. And does the AD operate in sync with the DA?

If not, then a 96KHz sampled tone at 40KHz has 2.4 samples per cycle. Each loop run through the AD can have the 40KHz tone with different 2.4 samples, and at high frequencies the values bounce around. So without perfect synchronization, it is like throwing a dice, the last run presents a new big data error. I see some of that problem with 44.1KHz sampling and 20KHz tone.

So does the software requires synchronization of input data to DA clock and AD?

 

[pkane]

Say some AD or DA takes advantage of the concept of noise shaping. It is a pretty old concept widely used. Say some noise shaping at 96KHz sample rate “moves” 5dB of the noise to 24KHz-48KHz frequency range. In fact, there is a slope with the level growing with increased frequency. That or any other high frequency energy is a problem.

If you run a loop 10 times, the noise will accumulate to 50dB. That will show up in the digital comparison as a huge problem. That noise is in the data, poor converter gets an F.

My second question: Does your software require to synchronize the data source to the DA clock. This is not a common AES or SPDIF DA, it will requires separate clock input to the DA. And does the AD operate in sync with the DA?

If not, then a 96KHz sampled tone at 40KHz has 2.4 samples per cycle. Each loop run through the AD can have the 40KHz tone with different 2.4 samples, and at high frequencies the values bounce around. So without perfect synchronization, it is like throwing a dice, the last run presents a new big data error. I see some of that problem with 44.1KHz sampling and 20KHz tone.

So does the software requires synchronization of input data to DA clock and AD?

1. A typical null test requires just one pass through the loop. Any accumulated noise produced by the loop will be apparent and measurable in the loopback, and this is as it should be. Noise shaping reallocates quantization noise out to (hopefully) less audible frequencies. A null analysis with a loopback will show added noise at those frequencies. Where's the problem? The noise was really added by the conversion at those frequencies, so it should be detected.

2. DeltaWave doesn't require DA/AD synchronization. Synchronization is done by the software to a tiny fraction of a sample. It also measures and removes clock drift if DA and AD clocks are not synchronized. As to the possibility of multiple matches, the software is designed to work with complex music recording, not a simple sine wave. While the possibility of multiple matches is extremely low with any non-digitally generated material, the software will detect this and alert the user should it happen.

 

[Dan Lavry]

1. A typical null test requires just one pass through the loop. Any accumulated noise produced by the loop will be apparent and measurable in the loopback, and this is as it should be. Noise shaping reallocates quantization noise out to (hopefully) less audible frequencies. A null analysis with a loopback will show added noise at those frequencies. Where's the problem? The noise was really added by the conversion at those frequencies, so it should be detected.

2. DeltaWave doesn't require DA/AD synchronization. Synchronization is done by the software to a tiny fraction of a sample. It also measures and removes clock drift if DA and AD clocks are not synchronized. As to the possibility of multiple matches, the software is designed to work with complex music recording, not a simple sine wave. While the possibility of multiple matches is extremely low with any non-digitally generated material, the software will detect this and alert the user should it happen.

It is, of course about the slew rate, the slope as you cross the 0V of a sine wave. The slew rate for full scale tone at 20KHz is around 125000 volts per second. 250000V/S.

Let me try to figure what it would take to design a 100dB hardware test system (far below Audio precision).

For a 1V peak sine wave V= 1* sin(w*t) a tone at -100dB level is 10uV. So an time error for 20KHz is 10uV/125000= 80 pico second

For 40KHz and 80KHz signal it is 40psec and 20psec.

Wow, 20psec is the limit of my test gear for jitter, so far so good, but what about slow time drift? The solution is obvious, an atomic clock. But what about a couple of degree temperature rise… And still there is jitter, usually much higher than 20psec. Pretty easy with a 50GHz computer clock speed, offering 20psec resolution and some real fast laptop of the future. No worry the software will do it. Somehow.

So take two files with some different data, say around 10uV and line them or compensate to 80psec accuracy for 20KHz tone? 2psec to measure 80Khz to 120dB test software? If I could do it in hardware, Elon would pay 100 million…

So the answer is “I do it in software”. A pico second is a millionth of a millionth of a second. Time determination of 80psec based on music that can be off by errors bigger than 10 parts per million.. Music for a long time still has many parts per million errors.

Wow. And all of that while “best fit” is different than correct alignment. I think of it as the pico second bridge.

That test gave my Savitr AD, my best design ever, poor rating. So either the software is broken, or the specific test of my gear was done wrong.

And here we already established that your software as is (without a filter) would consider energy above hearing as a flaw. So the software is already at best limited to 44.1KHz tests. To test 96KHz or 192KHz the user should implement some fantastic low pass filter at 20KHz. What is the point of testing 192KHz with 20KHz low pass? And even at 44.1KHz correcting for 80 psec requires equivalent time resolution of 12.5GHz?

There is more, but let’s put that “difficult to believe” aside for a moment.

I just wanted to see what people say about Savitr. The thing that came out of the search was bad rating. Nothing else, posted years ago. I guessed that anyone doing the same search would lose interest. The people on the thread are not likely customers. We don’t advertise and it is less scientifically oriented group. I was surprised, some guy hated me, I guessed I challenged his buying decision I felt un-welcomed to say the least. And pkane you were one of three leader of the pack, the main one, the guy that wrote the software.

I don’t want to ruin pkane’s day, but he helped ruined mine. He can fix it. I would at the minimum wish the acknowledgment on that gear space thread that either the software is always flawed, or there was some operator error (BIG error), for whatever reason. An apology with clear acknowledgement of the Savitr’s well deserved excellence, strong enough statement to compensate for the years of poor reputation.

I don’t like people reading poor rating on my excellent converter.

 

[pkane]

So the answer is “I do it in software”. A pico second is a millionth of a millionth of a second. Time determination of 80psec based on music that can be off by errors bigger than 10 parts per million.. Music for a long time still has many parts per million errors.

Dan, the precision of clock difference synchronization in software can be measured and analyzed. Remember, I'm not synchronizing each individual sample, I'm synchronizing about 5 to 10 million of them simultaneously, more if the music is longer than a few minutes or the sample rate is higher than 44.1k. Individual sample precision isn't important in the synchronization calculation.

And here we already established that your software as is (without a filter) would consider energy above hearing as a flaw.

Don't know how you decided this is a flaw. My software shows the spectrum and phase errors on a frequency plot. You decide if noise or errors above 20k is a flaw or not, it's not my decision. DeltaWave also shows a psychoacoustic-based analysis that uses audibility curves, known temporal and critical band masking to compute an audibility-weighted error metric. While I don't tell anyone that this is what they'll hear, it's designed as a guide to determine where the particular recording errors may become audible.

I just wanted to see what people say about Savitr. The thing that came out of the search was bad rating.

And pkane you were one of three leader of the pack, the main one, the guy that wrote the software.

What you apparently missed in that thread was that I wasn't defending the results of the test as is done by Didier, and he is not using my software.

I've been critical of Didier's results for quite a few years. A simple DC filter will alter phase enough to result in a 10-20dB difference in his rating. I've shown this flaw with DeltaWave by performing phase linearization in software, others have demonstrated it by removing or bypassing the coupling capacitors. It's clearly a problem in his test. The use of different reconstruction filters with different phase behavior in these tests make it even harder to compare the results between devices and measurements. Again, I'm in no way defending his results (or have anything to do with them!)

I don’t want to ruin pkane’s day, but he helped ruined mine. He can fix it.

I would at the minimum wish the acknowledgment on that gear space thread that either the software is always flawed, or there was some operator error (BIG error), for whatever reason.

Sorry, Dan, but you're still confused: I can't fix it, because I don't run the thread or do the measurements there, or provide the software that Didier uses. He refuses to make any changes to the test whenever I bring these issues up, and I've done it many times over the years, both, in the thread, and privately. I've done the work to eliminate the unimportant variables from null analysis in DeltaWave, but I can't make Didier do the same.

I've also mentioned in that thread that I took the loopback result with Savitr that was posted there, and ran it through my software using the phase linearization option. The result was an RMS error of around -100dB compared to what Didier measured: -56dB. Huge difference.

 

[Dan Lavry]

pkane, here is my point of view: I came to a forum looking to see what people say about my Savitr, last design 3 years effort, doing everything (concept, math, schematic, PC layout…). I was shocked to find out the terrible rating, and nothing else on that forum.

There are thousands of posts there going back many years. So I don’t know the background. And at 80 years old, I have little interest in learning how software works. What I saw was a free software test that gives my product bad marks. Once we agreed on the thread that AP does matter, I posted some Savitr test results (THD+N plot over 20-20KHz, -126dB noise, 0.0002% at 1KHz…

There was no acknowledgment that there is an issue here. The bad mark still there. Later on you said something about 100dB. The AP shows better than 100dB result. So what is that 100dB mean? I guess it is RMS like THD+N. And measuring with music? It yields worse result for loud high frequencies.

If I am not mistaken, 80KHz full scale has 500uV/second slew rate. So 1nsec time alignment ALONE yield -78dB tone at 20KHz, -72dB at 40KHz, -66dB at 80KHz. To get -100dB for 20KHz tone 1uV / 125000V/sec slew rate, the timing accuracy needed is 8psec, so I have my doubts. But that aside, it is very limiting for a test system. How does real music relate to it all? Using loud high frequency energy for more accurate results?

But it did not matter at the time. What I saw is years of posts with lists of product combinations in pairs. People waiting for a tester to tell them results. These is not a “normal studio”. A guy (maybe more than one) gets his hand on numerous equipment over 10 years, hooks them in pairs, different combinations, listing result numbers next to each pair, and my numbers were bad. I did see discussion about looping many times, or running just one pass.

At first, it looked to me like a possible commercial operation (access to numerous gear and time and energy to do it all). But that would involve some evil competitor. Or it could be the software. The thing I could solidly stand behind was to post some Savitr test result (post #3493). Everything else is like hearsay on the internet.

So I admit that I was not open to much conversation, and possibly not nice to deal with. But how did you reply? Your post #3503 shows an FFT of DA AD pair with third harmonic at -140dB, 2nd 4th and 5th at -150dB. So much better than Savitr or AP. I normally would like to see a 5-10KHz but I first I want to know why you refuse to say the name and model of the converter to stand behind your FFT. I asked more than once.

I complained that the FFT shows a total of 0.00001% distortions. So the next FFT (post #3510) shows another 1KHz tone with a 3rd at -135dB. One more mysterious unit with great results. Again, repeated requests for this and the $700 magic DA/AD names or models is silence. This is not fair and un professional. In contrast with my specifications offering more details, presented in the open and can be verified.

I was there with a valid complain. we were there almost every day for at least a couple of weeks. But when I talked about problems with multiple loops, you did not suggest that the Savitr bad results (and a few other companies complain in the background) could be due to using more than one pass, as you said now. When I talked about the issue of a poor unit in series with good one, you did not say anything about the test requiring a real good AD (ironically like the Savitr). To me it seems like you joined the crowd. A couple of people declared that I am a loser, I got a hate massage.

Now you say that you can’t fix the problem, because you don’t run the forum or web. But you can post a massage. Posting “stand alone” FFT’s from unknown product is not very professional to say the least. So for me you became part of the nonsense crowd. From my point of view, here is your opportunity to fix it, assuming you care about fairness and justice..

I don’t expect you to reveal the model of the $700 combo, but you can certainly say that looping multiple times is unwise, and about testing a pair using a top notch like Savitr, and more. I am not sure that the group react to one pass or using good quality reference device, or comments about the numerous lists of converter pairs. You don’t have to say any of it. But you posted the mysterious FFT’s. I think you should look at my test results (#3493) and acknowledge that this is an exceptional performance for audio AD. We both know it. You seem to agree NOW that the evaluation was flawed. I did not need to do learn about software. I want the people on this thread to hear it from you, saying that rating Savitr low was truly unfair. My focus is the reputation of Savitr, three years effort resulting in my best accomplishment. You are not totally free from any connection with the people on that thread attitude towards my product and me as a person.

 

[pkane]

pkane, here is my point of view: I came to a forum looking to see what people say about my Savitr, last design 3 years effort, doing everything (concept, math, schematic, PC layout…). I was shocked to find out the terrible rating, and nothing else on that forum.

There are thousands of posts there going back many years. So I don’t know the background. And at 80 years old, I have little interest in learning how software works. What I saw was a free software test that gives my product bad marks. Once we agreed on the thread that AP does matter, I posted some Savitr test results (THD+N plot over 20-20KHz, -126dB noise, 0.0002% at 1KHz…

There was no acknowledgment that there is an issue here. The bad mark still there. Later on you said something about 100dB. The AP shows better than 100dB result. So what is that 100dB mean? I guess it is RMS like THD+N. And measuring with music? It yields worse result for loud high frequencies.

If I am not mistaken, 80KHz full scale has 500uV/second slew rate. So 1nsec time alignment ALONE yield -78dB tone at 20KHz, -72dB at 40KHz, -66dB at 80KHz. To get -100dB for 20KHz tone 1uV / 125000V/sec slew rate, the timing accuracy needed is 8psec, so I have my doubts. But that aside, it is very limiting for a test system. How does real music relate to it all? Using loud high frequency energy for more accurate results?

But it did not matter at the time. What I saw is years of posts with lists of product combinations in pairs. People waiting for a tester to tell them results. These is not a “normal studio”. A guy (maybe more than one) gets his hand on numerous equipment over 10 years, hooks them in pairs, different combinations, listing result numbers next to each pair, and my numbers were bad. I did see discussion about looping many times, or running just one pass.

At first, it looked to me like a possible commercial operation (access to numerous gear and time and energy to do it all). But that would involve some evil competitor. Or it could be the software. The thing I could solidly stand behind was to post some Savitr test result (post #3493). Everything else is like hearsay on the internet.

So I admit that I was not open to much conversation, and possibly not nice to deal with. But how did you reply? Your post #3503 shows an FFT of DA AD pair with third harmonic at -140dB, 2nd 4th and 5th at -150dB. So much better than Savitr or AP. I normally would like to see a 5-10KHz but I first I want to know why you refuse to say the name and model of the converter to stand behind your FFT. I asked more than once.

I complained that the FFT shows a total of 0.00001% distortions. So the next FFT (post #3510) shows another 1KHz tone with a 3rd at -135dB. One more mysterious unit with great results. Again, repeated requests for this and the $700 magic DA/AD names or models is silence. This is not fair and un professional. In contrast with my specifications offering more details, presented in the open and can be verified.

I was there with a valid complain. we were there almost every day for at least a couple of weeks. But when I talked about problems with multiple loops, you did not suggest that the Savitr bad results (and a few other companies complain in the background) could be due to using more than one pass, as you said now. When I talked about the issue of a poor unit in series with good one, you did not say anything about the test requiring a real good AD (ironically like the Savitr). To me it seems like you joined the crowd. A couple of people declared that I am a loser, I got a hate massage.

Now you say that you can’t fix the problem, because you don’t run the forum or web. But you can post a massage. Posting “stand alone” FFT’s from unknown product is not very professional to say the least. So for me you became part of the nonsense crowd. From my point of view, here is your opportunity to fix it, assuming you care about fairness and justice..

I don’t expect you to reveal the model of the $700 combo, but you can certainly say that looping multiple times is unwise, and about testing a pair using a top notch like Savitr, and more. I am not sure that the group react to one pass or using good quality reference device, or comments about the numerous lists of converter pairs. You don’t have to say any of it. But you posted the mysterious FFT’s. I think you should look at my test results (#3493) and acknowledge that this is an exceptional performance for audio AD. We both know it. You seem to agree NOW that the evaluation was flawed. I did not need to do learn about software. I want the people on this thread to hear it from you, saying that rating Savitr low was truly unfair. My focus is the reputation of Savitr, three years effort resulting in my best accomplishment. You are not totally free from any connection with the people on that thread attitude towards my product and me as a person.

Dan, I understand where you're coming from, and the reason the conversation went sideways on gear space. I'm not in charge of anything there and can't fix anything that Didier does, but for some reason you decided that I am and that I was defending his results. Far from it. I did attempt to explain how a (proper) null test works, and why the results posted there are misleading, but we were talking past each other, apparently.

did not need to do learn about software. I want the people on this thread to hear it from you, saying that rating Savitr low was truly unfair.

Sure, I said it before. I specifically said there that the result is largely skewed by DC filter and the reconstruction filter phase response, and how to fix the test so it's not so affected. This is something I've posted multiple times before in that thread, and also in a direct answer to you. I assume you missed it or misread it. Years ago I went so far as to re-run most of Didier's tests and posted the comparison using phase linearization to show how far off his results really are. I'm not sure I can (or want) to do much more than that.

If you really want, we can continue this in a private conversation as I'm not sure anyone here is interested in all the gory details or the history of that other thread.

 

[Dan Lavry]

I will make it shorter, you tried to distance yourself but you were involved.

I was compelled to go to the lab, take screen shots… post some Saviter test results. I wanted to make my point clear. (post#3493), April 8.

I don’t know why, but NEXT DAY April 9 you posted (post#3503) a 1KHz FFT along with the statement: “You can decide for yourself if the AP or the amazing Savitr are capable of this level of performance (total cost for DA + AD of around $700). I'm out of this conversation”. The pot is made with average of 32 plots to reduce the noise far enough. Better than Savitr by 20dB.

It looked like feeding a Savitr -20dB tone and doing 32 avarages, but of course it was not. The top of the signal shown is at 0dB not -20dB, but still enough to show that something is wrong. The magic $700 unit is unknown.

You were not out of the conversation. You placed another 1KHz FFT later but refuse to the day to tell the model or make of that wonderful DA AD pair.

So you “showed them” that Lavry is no big deal. So as far as I am concerned you did the dirty laundry part, and I am getting all the negativity in that thread. In my opinion you owe me to fix the problem.

I don't do dirty laundry and I like day light. I am not ready to communicate directly. You are the guy that put the weird FFT of still unknown source, trying or helping to minimize the product and the designer. You know what you did, you know the Savitr is great. You can take responsibility or not. The ball is in your court.

 

[pkane]

I will make it shorter, you tried to distance yourself but you were involved.

I was compelled to go to the lab, take screen shots… post some Saviter test results. I wanted to make my point clear. (post#3493), April 8.

I don’t know why, but NEXT DAY April 9 you posted (post#3503) a 1KHz FFT along with the statement: “You can decide for yourself if the AP or the amazing Savitr are capable of this level of performance (total cost for DA + AD of around $700). I'm out of this conversation”. The pot is made with average of 32 plots to reduce the noise far enough. Better than Savitr by 20dB.

It looked like feeding a Savitr -20dB tone and doing 32 avarages, but of course it was not. The top of the signal shown is at 0dB not -20dB, but still enough to show that something is wrong. The magic $700 unit is unknown.

You were not out of the conversation. You placed another 1KHz FFT later but refuse to the day to tell the model or make of that wonderful DA AD pair.

So you “showed them” that Lavry is no big deal. So as far as I am concerned you did the dirty laundry part, and I am getting all the negativity in that thread. In my opinion you owe me to fix the problem.

I don't do dirty laundry and I like day light. I am not ready to communicate directly. You are the guy that put the weird FFT of still unknown source, trying or helping to minimize the product and the designer. You know what you did, you know the Savitr is great. You can take responsibility or not. The ball is in your court.

Sorry, Dan. Not rehashing it here. As I said, if you want to continue the conversation, message me privately.

 

[Dan Lavry]

Sorry, Dan. Not rehashing it here. As I said, if you want to continue the conversation, message me privately.

It does not need rehashing here. It needs to be corrected there at the thread. Your refusal to correct your action (The LET'S and unknown magic $700 DAAD in response to the Savitr) is not a reflection on me. I have credibility.

 

[Dan Lavry]

Officially, at 12p.m. and 1am I will be a year older. That, of course has no meaning in the physical world. J.W. Gibbs explained observation in 1899. The reality is that “you can’t do anything in zero time”. Say the goal is to change from 0 to 1 a “clean” 90-degree step. The real-world outcome starts at 0, but it takes time to “clime up” (rise time). The signal passes 1.00 and “climbs” to 1.09 (9% overshoot) the first cycle of decayed ringing. 1KHz or 1GHz bandwidth it is always around 9%. The ringing frequency is bandwidth cutoff frequency. The decay is hyperbolic v= Sin(w*t)/t (multiply by 1/t).

Sine waves: A simple single step is enough to bring up that sine wave (sinusoidal). Nature is full of waves, harmonic behavior made out of “elements”. Much of sound contains harmonic energy, more so for music instruments designed to generate harmonics.

One can reduce the impact of Gibbs phenomena and all signal continuities with “windowing”, rounding the sharp corners. One more time, the sine wave shape shows up in constructing the windows.

We should know by now that much of sound energy (air vibration and music energy in particular can be viewed as made out of “components”, each component is defined by frequency, amplitude and phase. Simple addition of the “components” is the same as the “original”.

That is the fundamental idea behind audio testing, to measure how the device behaves when encountered by each sine wave component. If it behaves well for a tone, it will do so for music. But music does not behave like a constant frequency tone. Two or more “unrelated” sine wave components are used to probe possible intermodulation distortions, (should be under the noise floor). For example, a -12dB 7KHz “riding” on 0dB 60Hz yields a constantly changing wave. We know that the tones should not interact, and where to look for distortions.

Sine wave-based testing is the way to test audio (like Audio Precision) My job, as a designer and a manufacturer is to ensure that the AD can accommodate all possible music by checking cases, how the unit treats all the components. Clearly “all” cases are not practical, but say 100 frequencies from 20-20KHz, spaced linearly or logarithmically. Or say -100-point plot of 0dB-120dB level change…

I read opinions that a system based on sine waves can’t test “real music”. I am all for good listening tests (set up properly). But “exercising” a converter with various music in a quite facility is a “big deal”. It takes time attention, and “operator” and two or more good ears to listen different music tracks say all in agreement. But with music the information is “soft”. There is always a question about listening to enough variety and more.

The test system can force, measure and quantify all that is needed. For example, a full-scale tone at 20KHz or more. If the AD operating correctly for such tone, it will behave well when 20KHz is 5th harmonic of 4KHz, or 4th harmonic of 5KHz… The system enables probing the complete range of human hearing using sine wave tones and combinations.

Again, audio testing with sine waves and combinations enables probing the whole range of the components that make up the signal (music). Separating the sound into components (such as fundamental and harmonics, phase, flatness response). It enables separation of parameters such as amplitude-frequency, THD+N, noise…

I lot of people don’t believe in measurements due to little specification details or other reasons. I understand that a product for the ear should be evaluated by the ear. I design for the ear. Needless to say I am very glad when some tests my AD with an Audio Precision. I am very pleased when successful pro audio people use my AD to spread good music around the world. Needless to say there is a lot more to AD design for for use in professional studio environment.

Back to my birthday and Gibbs: It is almost 10p.m. I need to start ramping up to be at 50% at 12p.m. I hope the pace maker does not ring at 2a.m tomorrow.

 

[MaxwellsEq]

Officially, at 12p.m. and 1am I will be a year older. That, of course has no meaning in the physical world. J.W. Gibbs explained observation in 1899. The reality is that “you can’t do anything in zero time”. Say the goal is to change from 0 to 1 a “clean” 90-degree step. The real-world outcome starts at 0, but it takes time to “clime up” (rise time). The signal passes 1.00 and “climbs” to 1.09 (9% overshoot) the first cycle of decayed ringing. 1KHz or 1GHz bandwidth it is always around 9%. The ringing frequency is bandwidth cutoff frequency. The decay is hyperbolic v= Sin(w*t)/t (multiply by 1/t).

Sine waves: A simple single step is enough to bring up that sine wave (sinusoidal). Nature is full of waves, harmonic behavior made out of “elements”. Much of sound contains harmonic energy, more so for music instruments designed to generate harmonics.

One can reduce the impact of Gibbs phenomena and all signal continuities with “windowing”, rounding the sharp corners. One more time, the sine wave shape shows up in constructing the windows.

We should know by now that much of sound energy (air vibration and music energy in particular can be viewed as made out of “components”, each component is defined by frequency, amplitude and phase. Simple addition of the “components” is the same as the “original”.

That is the fundamental idea behind audio testing, to measure how the device behaves when encountered by each sine wave component. If it behaves well for a tone, it will do so for music. But music does not behave like a constant frequency tone. Two or more “unrelated” sine wave components are used to probe possible intermodulation distortions, (should be under the noise floor). For example, a -12dB 7KHz “riding” on 0dB 60Hz yields a constantly changing wave. We know that the tones should not interact, and where to look for distortions.

Sine wave-based testing is the way to test audio (like Audio Precision) My job, as a designer and a manufacturer is to ensure that the AD can accommodate all possible music by checking cases, how the unit treats all the components. Clearly “all” cases are not practical, but say 100 frequencies from 20-20KHz, spaced linearly or logarithmically. Or say -100-point plot of 0dB-120dB level change…

I read opinions that a system based on sine waves can’t test “real music”. I am all for good listening tests (set up properly). But “exercising” a converter with various music in a quite facility is a “big deal”. It takes time attention, and “operator” and two or more good ears to listen different music tracks say all in agreement. But with music the information is “soft”. There is always a question about listening to enough variety and more.

The test system can force, measure and quantify all that is needed. For example, a full-scale tone at 20KHz or more. If the AD operating correctly for such tone, it will behave well when 20KHz is 5th harmonic of 4KHz, or 4th harmonic of 5KHz… The system enables probing the complete range of human hearing using sine wave tones and combinations.

Again, audio testing with sine waves and combinations enables probing the whole range of the components that make up the signal (music). Separating the sound into components (such as fundamental and harmonics, phase, flatness response). It enables separation of parameters such as amplitude-frequency, THD+N, noise…

I lot of people don’t believe in measurements due to little specification details or other reasons. I understand that a product for the ear should be evaluated by the ear. I design for the ear. Needless to say I am very glad when some tests my AD with an Audio Precision. I am very pleased when successful pro audio people use my AD to spread good music around the world. Needless to say there is a lot more to AD design for for use in professional studio environment.

Back to my birthday and Gibbs: It is almost 10p.m. I need to start ramping up to be at 50% at 12p.m. I hope the pace maker does not ring at 2a.m tomorrow.

This thread blew up out of nowhere after more than 5 years of silence. And got a bit heated. I tried to understand what was going on.

It is clear from the thread linked here, that whilst using a null is valuable, it has limitations: https://www.audiosciencereview.com/...c-filter-effect-on-the-null-rms-metric.61930/

Creating a single number from a null is even more fraught. This is especially true if the null does not account for differences in DC-blocking and high frequency roll-off, especially the fs/2 "brickwall filter" behaviour (if a DAC or ADC is under test)

I feel these experimental limitations undermine any useful ranking conclusions that can be drawn from the Gearspace method using "diffmaker" (not Deltawave) software. I'm not sure anyone on ASR would defend the Gearspace single-number null approach.

 

[danadam]

This thread blew up out of nowhere after more than 5 years of silence. And got a bit heated. I tried to understand what was going on.

The messages were moved here from:

Master Thread: Are Measurements Everything or Nothing?

Yes, Dan, but nobody is claiming that the multigenerational loop-back test measures the specific performance of the converters. They are constructing a demonstration to show that the faults of converters are inaudible, if the first conversion measures well (flat response, negligible distortions...

www.audiosciencereview.com

 

[MaxwellsEq]

The messages were moved here from:

Master Thread: Are Measurements Everything or Nothing?

Yes, Dan, but nobody is claiming that the multigenerational loop-back test measures the specific performance of the converters. They are constructing a demonstration to show that the faults of converters are inaudible, if the first conversion measures well (flat response, negligible distortions...

www.audiosciencereview.com

Ahhh! That explains the apparent lack of coherence