SINAD vs. $

[pozz]

The SINAD vs. $ question.

Edit: The graphs below were made before there was a proper database or an interactive table. For the current, complete dataset please see the Attribute Comparison Tool for electronics and speakers by @RickSanchez.

The scatterplots show that all measured hardware is between 40dB—120dB SINAD. Putting in a $10k limit, past about $3k—$5k there is a clear downwards trend, and before then there's somewhat even dispersion, although that's only because the values tend to occupy the range between 80dB—110dB.

I think, given more expensive entries, the graph will largely populate sideways.

Without labels:

With labels:

 

[Ron Texas]

It would be nice if the legibility could be improved on the graphs with labels, especially the $1000 limit. I wonder what it would look like instead of just price, the amplifiers could be ranked by dollars per watt on the horizontal axis, or just watts with price ranges indicated by text color.

 

[pozz]

I can make one for amps with $ vs. watts, but I won't make ranked charts. Many people will miss the point.

 

[GrimSurfer]

This is an excellent depiction, @pozz.

My take-away is that there is no absolute correlation between SINAD and $s. There is some evidence of an inverse relationship, but this only becomes apparent for equipment at the opposite ends of the price spectrum.

The small sampling numbers for gear in the upper half of the price range increases the margin of error of any assessed relationship between SINAD and price.

 

[garbulky]

You could also use audibility and price put that at say 80 db or 75 db. Then show the value range for audibility in measurements. You'll find that everything over $50 or $100 is simply going to be adding more $ even if it's bumping the SNR value.

 

[617]

I was looking at this data the other day although my analysis is not as sophisticated. It seems like the correlation between cost and SINAD is not strong, but there are distinct plateaus in SINAD.

 

[GrimSurfer]

I was looking at this data the other day although my analysis is not as sophisticated. It seems like the correlation between cost and SINAD is not strong, but there are distinct plateaus in SINAD.

It would be interesting to calculate standard deviation from the mean. This could then be used to greater effect than, say, comparing an expensive POS to a well performing dongle.

 

[pozz]

You could also use audibility and price put that at say 80 db or 75 db. Then show the value range for audibility in measurements. You'll find that everything over $50 or $100 is simply going to be adding more $ even if it's bumping the SNR value.

I'm not sure what you mean, but here it is. Up to $500 you'll find everything.

Also, it's SINAD, not SNR.

 

[pozz]

It would be interesting to calculate standard deviation from the mean. This could then be used to greater effect than, say, comparing an expensive POS to a well performing dongle.

This is what that looks like:

Edit: The APx555 was not included in the set.

 

[617]

It would be nice if the data from these reviews was tabulated. It might be nice to search by feature.

 

[GrimSurfer]

This is what that looks like:

View attachment 35566

Excellent!

The standard deviation for price is being skewed by the $28k item (APx555). If you drop that data point $s and SINAD) from all calculations, the result will be less skewed and more representative of the whole.

(Besides, it might be a bit of a stretch to include the performance of a spectrum analyzer in a comparison of consumer audio components... as much as I'd love to have that kind of precision in my audio rack.)

 

[Juhazi]

This would look a lot nicer with log scales

 

[KSTR]

IME, SINAD is not a valid/sufficent measure of audio quality. It's a measured quantity, and a very crude and condensed one at that. And retail prize is not mainly influenced by design effort, technical competence and flawless execution. So I would think those scatter plots have no deeper value to them than, well, we can see that SINAD does not correlate much with cost.

Nonetheless, thanks @pozz for compiling.
To improve the plots I suggest to use log scale on the money axis as well.
EDIT: Ah, @Juhazi just had the same idea ;-)

 

[pozz]

It would be nice if the legibility could be improved on the graphs with labels, especially the $1000 limit. I wonder what it would look like instead of just price, the amplifiers could be ranked by dollars per watt on the horizontal axis, or just watts with price ranges indicated by text color.

Here are scatterplots, mean and StDev.

Edit: Reposted after finding a calculation error.

 

[pozz]

So I would think those scatter plots have no deeper value to them than, well, we can see that SINAD does not correlate much with cost.

That's pretty much it.

$ log scaled.

 

[KSTR]

Hey, you're quick and know your tools ;-)

 

[pozz]

Excellent!

The standard deviation for price is being skewed by the $28k item (APx555). If you drop that data point $s and SINAD) from all calculations, the result will be less skewed and more representative of the whole.

(Besides, it might be a bit of a stretch to include the performance of a spectrum analyzer in a comparison of consumer audio components... as much as I'd love to have that kind of precision in my audio rack.)

Sorry, I should have mentioned that I removed the AP analyzer from the set. So the skewness is largely due to the Totaldac and the Linn streamer.

Edit: Removing the Totaldac d1-six and Linn Akurate DSM affects the stats like this:

 

[GrimSurfer]

Challenging having such large StdDevs, as it removes a great deal of meaning and subtlety from the calculation.

 

[DonH56]

I have found over the years that many high-end components (price-wise) have worse technical performance than cheaper (make that "less-expensive") products. My total hand-waving thoughts over the years have been that it is a combination of some very expensive designs being driven by folk with little design/engineering expertise and the desire and effort to achieve a certain sound based upon their listening tests rather than an absolute standard for accuracy or whatever. A number of high-end vendors actually flaunt their lack of engineering in favor of listening. The listening aspect so revered by many (been there!) has the problem of a tendency to optimize for the system used for the assessment. That can lead to a "one horse (system) wonder", i.e. when an amplifier is tailored to sound right driving a specific speaker and so forth, it might not do so well on others. And of course things like MHz audio amps that need special cables to filter out HF noise and keep them stable driving difficult speaker loads.

Onwards - Don

 

[GrimSurfer]

Very interesting view, Don. I wonder if that gave rise to the speaker matching/Golden Ear phenomena?