Speed Reading DAC Audio Measurements

[Boxermotor]

Thank you so much everyone. What would you like the next speed run video be about? I am thinking of either headphone amplifiers or speakers.

Would enjoy either, but would especially like to see how you evaluate speakers. I'm in the process of deciding on what to use for a near field setup and have been intrigued by your glowing endorsement of the Revel M106's (objectively and subjectively) and how your measurements indicate the objective quality of a speaker.

 

[phoenixdogfan]

Keyword here is that SINAD is good to PREDICT the overall performance of the DAC, not that is an end all be all of measurements. The remaining ones give a better overall picture but Amir nails it: when SINAD is very high, very likely all the rest is excellent as well.

In economics we call that a proxy--a measure which while not completely conclusive is nonetheless highly likely to yield a correct assessment of the overall state of the system.

 

[beeface]

Thank you so much everyone. What would you like the next speed run video be about? I am thinking of either headphone amplifiers or speakers.

Speaker measurements, personally, but either would be much appreciated.

 

[DonR]

Another speaker vote here

 

[amirm]

In economics we call that a proxy--a measure which while not completely conclusive is nonetheless highly likely to yield a correct assessment of the overall state of the system.

Ah, good word! I am going to steal it for future discussions.

 

[jasonhanjk]

Enob is multiply by 6.02, to be exact.

 

[amirm]

Enob is multiply by 6.02, to be exact.

Actually you have to first subtract 1.76 dB and then do that divide! Speed reading though calls for approximation and hence my rule of thumb.

 

[Rja4000]

Good explanation.

By the way: on the 32 tones plot, only 31 tones are shown, not 32 as you say in the video.
Which is normal, since the test signal includes a 14Hz tone, that doesn't show on the 20Hz-20kHz plot.
A detail, of course.

 

[DWPress]

Awesome Amir.

With ASR growing and continually hitting Google top 5 in any "audiophile" search this is a GIFT to those interested in the science of sound and home audio. I joined 5 years ago and it what a struggle early on understanding everything presented to me. I was already doing active XO, DSP, speaker building so familiar with measurements, graphs etc. but just interacting with fellow enthusiasts, developers and builders here on ASR (and lots of recommended reading) has taken things to a whole new level.

So, in a nutshell, thank you @amirm, wish you'd thought of it 5 years ago but this will help a lot of people.

 

[pavuol]

You are pronouncing SINAD wrong. Sy-Nad is wrong. Sin-Ad is right. :P

and that istn't the worst part, just when you turn subtitles on you realize Amir acts positive but talks toxic!

Spoiler: shinAYD synad Sinéad ..err, whatever :)

I could now rant how the ALMIGHTY AI isnt't able to transcript a single term when the context was clearly set in previous speech but hey, maybe it knows VERY well what it does and tells us the old truth - that this whole hobby/industry is toxic

 

[ZgSTar]

Great video. Thanks

 

[antcollinet]

Excellent video, clarifiying one or two points for me (I'm looking forward to the speaker measurement explantion which have have a very poor feel for)

One question - around 4:40 - "This is not the real noise floor" I"ve heard this before when applied to FFT measurements, but don't understand it.

If it is not real noise floor, what is?
What is it about the measurement that doesn't show the real noise floor.

Anyone?
Thanks.

 

[antcollinet]

Thank you so much everyone. What would you like the next speed run video be about? I am thinking of either headphone amplifiers or speakers.

Speakers please. Much of those lack meaning for me.

 

[amirm]

One question - around 4:40 - "This is not the real noise floor" I"ve heard this before when applied to FFT measurements, but don't understand it.

If it is not real noise floor, what is?
What is it about the measurement that doesn't show the real noise floor.

With FFT we get to take the noise floor and spread it over many "buckets" (FFT length). The more buckets we use, the lower the displayed noise. To know the actual noise floor, you need to know the FFT length. Using simple math, you convert that to dB and add it to the displayed value. Different FFTs in my measurements have different number of points. I think the one in the dashboard lowers the noise floor by 35 dB or so (from memory). So you need to add that to what you see to get the actual noise floor. This is called "FFT gain." A kind of free lunch in signal processing to allow us to dig deep below the noise floor to find actual signals. Our hearing by the way has similar ability to tune of 5 to 10 dB or so.

Instead of using the above math, you can just look at the dynamic range measurements. That number is real.

 

[antcollinet]

With FFT we get to take the noise floor and spread it over many "buckets" (FFT length). The more buckets we use, the lower the displayed noise. To know the actual noise floor, you need to know the FFT length. Using simple math, you convert that to dB and add it to the displayed value. Different FFTs in my measurements have different number of points. I think the one in the dashboard lowers the noise floor by 35 dB or so (from memory). So you need to add that to what you see to get the actual noise floor. This is called "FFT gain." A kind of free lunch in signal processing to allow us to dig deep below the noise floor to find actual signals. Our hearing by the way has similar ability to tune of 5 to 10 dB or so.

Instead of using the above math, you can just look at the dynamic range measurements. That number is real.

Excellent, thanks - A bit more background reading based on that, and I'll probably get it.

 

[xaviescacs]

To deepen in the meaning of SINAD being a predictor we would need a number to represent other quantities like linearity and multi tone distortion and then try to build a very simple model for each one to predict it one from the SINAD. What amir means is that such models for each of the other measurments (represented by a real number) could be build and will work in most of DACs. The degree of linearity of each model would depend on how the summarizing variables of other measurements are defined. I personally like the use of the word "predictor" here.

 

[Purité Audio]

Superb, thank you A.
Keith

 

[Triliza]

Thank you, very useful video. You know better how many new users and visitors read the reviews, it may be useful for them to have an (unobtrusive) link on each review to these kind of videos for each category, assuming you make one for each category of the products you test (ninja style of saying please do so ).

 

[roog]

Thank you @amirm, for another great video, I found it very helpful to my understanding of your reviews. It is amusing to me that I have picked up a lot along the way just by reading your reviews but it really helps that you have explained the relative importance of parameters in the various test results in your video.

I would be grateful and of interest to me, if you would consider producing similar videos for analogue circuitry such as occur in pre amps and power amps, perhaps giving us an insight into the reasons why the artifacts we see in the test results occur.

Many thanks

 

[KSTR]

To deepen in the meaning of SINAD being a predictor we would need a number to represent other quantities like linearity and multi tone distortion and then try to build a very simple model for each one to predict it one from the SINAD. What amir means is that such models for each of the other measurments (represented by a real number) could be build and will work in most of DACs. The degree of linearity of each model would depend on how the summarizing variables of other measurements are defined. I personally like the use of the word "predictor" here.

1kHz SINAD / THD+N is not a useful predictor, especially not for IMD.
One can design two almost identical amps with the same superb 1kHz THD (THD, not SINAD or THD+N). Let one of them have 20dB less loop gain for correction at 20kHz and it will have much worse IMD.

Best universal test signal which allows much more insight than 1kHz single tone is a 7kHz+13kHz twin tone when measured with a 22kHz bandwidth, in an IMD (not IMD+N) analysis.

Lumping noise and distortion into one single value is a bad idea and always has been.

IMHO and IME, there are no shortcuts to a complete technical and mathematical understanding of what's going on when interpreting measurements. Speed reading and jumping to quick conclusions based on a single number is bound to fail.