After some more fiddling about, and some down time do to display failure, personal results are consistent with first impressions. Only a couple of the drivers are toss outs as implemented, the others all seem close; however one just seems to have more satisfying attack. I'm going to do more listening before casting vote.
Indeed some respond to this test by calling the Stereophile measurements a joke.
quote
"It depends on what measurements you see of the ncore.
Here are some measurements of an independent amplifier using NC1200 modules.
Theta Digital Prometheus monoblock power amplifier Measurements | Stereophile.com
They clearly do not measure as well as Hypex's datasheet would have you believe, but that could simply be down to implementation of the amplifier under review. Still for the price you're paying I would expect the implementation to be flawless."
quote
"It depends on what measurements you see of the ncore.
Here are some measurements of an independent amplifier using NC1200 modules.
Theta Digital Prometheus monoblock power amplifier Measurements | Stereophile.com
They clearly do not measure as well as Hypex's datasheet would have you believe, but that could simply be down to implementation of the amplifier under review. Still for the price you're paying I would expect the implementation to be flawless."
I asked the same thing earlier and the beginning of post 140, 5th element replied that it is an ASUS Xonar Essence ST analog inputs on the sound card. Measurement of which are performed by stereophile shown here:
ASUS Xonar Essence ST/STX soundcards Measurements | Stereophile.com
Stereophile's measurements are anything but a joke, they use an Audio Precision SYS2722 to perform all of their tests. AP stuff is pretty much an industry standard when it comes to reliability and consistency.
The only thing anyone could criticise would be John Atkinson's own comments regarding the results he gets, but the results speak for themselves.
Read comments below measurements. Performing measurements not according to accepted standards is OK, but don't try to say they aren't as good as Hypex data, they are uncomparable. One could say the Stereophile measurements are a joke because not according to standard, useless for comparison.
What comments below measurement are you talking about that says not to performed according to a certain standard? All I see are Atkinson's comments that it is superb even though it is a class D output.
This is getting a little OT so let's not go on much longer about amp measurements as this thread is about driver blind test listening.
Are you saying that your initial impressions sans DRC convolution are now coming back as the favored ones?
Maybe but from what I can tell those issues would more affect the small signal analysis and the frequency extremes.
Comparing the '8 ohm vs output power' between stereophile and hypex reveals two fairly similar graphs. Stereophile measures lower distortion at the very low power levels with hypex managing to reach the lowest minimum between 5 and 10 watts. The general trend and distortion levels above this are pretty much identical, with the differences probably accountable to the difference in rail voltages that the devices are being run from.
Certainly reaching 0.000x% distortion is no trivial task and simply mis-routing one wire inside the case would be plenty enough to ruin the hypex otherwise excellent performance. Or measuring with a wider measurement bandwidth.
For the frequency response vs distortion test, yes stereophile do use a larger bandwidth than hypex but there is very good reason for this. If you look at hypex graphs for the 100 watt and 250watt into 8 ohms you will see that at just under 7kHz the distortion rise, that was present before, quickly falls away. Why is this. Is it because the distortion actually disappeared? No quite the opposite, the distortion is still present and it's mainly third harmonic in nature, it disappeared because the measurement bandwidth that hypex use is limited to 20kHz. You physically stop being able to measure the third harmonic of such a measurement at 6.67khz, so no wonder it vanishes.
If you do not bandwidth limit the measurement to 20kHz then you are able to see the third harmonic in all its glory. This is what stereophile did. Sure increasing the measurement bandwidth also lets in noise so that the absolute distortion minimum is increased, but this is worthwhile doing for it lets you see exactly what the high frequency distortion performance of the amplifier is actually like. If you use a 20kHz limit then you're effectively sweeping the problem under the rug.
Achieving low distortion at low frequencies is easy, technically speaking no one cares about this (unless you do a really awful job), what's really telling of an amplifiers performance is how it does at high frequencies (5-20kHz). This is where the type of compensation the amplifier used becomes relevant and how the amplifiers stability margins and use of feedback were balanced - this is the interesting stuff. If you bandwidth limit to 20kHz all of this is lost.
It's lost in hypex own measurement TI do this too with most of their class D stuff because it is more becoming of a products performance and most people don't care to realise what bandwidth limiting this measurement actually means.
Stereophile's 80kHz measurements of the hypex vs frequency are most telling, because as you can see, the distortion rises very rapidly as frequency goes up. Hypex own measurements hint at this but you are not told the full story. This is what happens when the amount of feedback you can use is limited by stability margins and how an amplifier does this internally is particularly interesting. Clearly the hypex curtails its high frequency feedback very quickly.
TI, hats off to them, actually show you the difference between the two in their datasheet for their new TPA3251D2 class D single chip solution.
http://www.ti.com/lit/ds/symlink/tpa3251d2.pdf
You can see this here on page 10.
Figure one gives a graph that looks remarkably similar to hypex own datasheet, but figure 2 looks far more like stereophiles. Figure 2 is much more interesting.
Comparing the '8 ohm vs output power' between stereophile and hypex reveals two fairly similar graphs. Stereophile measures lower distortion at the very low power levels with hypex managing to reach the lowest minimum between 5 and 10 watts. The general trend and distortion levels above this are pretty much identical, with the differences probably accountable to the difference in rail voltages that the devices are being run from.
Certainly reaching 0.000x% distortion is no trivial task and simply mis-routing one wire inside the case would be plenty enough to ruin the hypex otherwise excellent performance. Or measuring with a wider measurement bandwidth.
For the frequency response vs distortion test, yes stereophile do use a larger bandwidth than hypex but there is very good reason for this. If you look at hypex graphs for the 100 watt and 250watt into 8 ohms you will see that at just under 7kHz the distortion rise, that was present before, quickly falls away. Why is this. Is it because the distortion actually disappeared? No quite the opposite, the distortion is still present and it's mainly third harmonic in nature, it disappeared because the measurement bandwidth that hypex use is limited to 20kHz. You physically stop being able to measure the third harmonic of such a measurement at 6.67khz, so no wonder it vanishes.
If you do not bandwidth limit the measurement to 20kHz then you are able to see the third harmonic in all its glory. This is what stereophile did. Sure increasing the measurement bandwidth also lets in noise so that the absolute distortion minimum is increased, but this is worthwhile doing for it lets you see exactly what the high frequency distortion performance of the amplifier is actually like. If you use a 20kHz limit then you're effectively sweeping the problem under the rug.
Achieving low distortion at low frequencies is easy, technically speaking no one cares about this (unless you do a really awful job), what's really telling of an amplifiers performance is how it does at high frequencies (5-20kHz). This is where the type of compensation the amplifier used becomes relevant and how the amplifiers stability margins and use of feedback were balanced - this is the interesting stuff. If you bandwidth limit to 20kHz all of this is lost.
It's lost in hypex own measurement TI do this too with most of their class D stuff because it is more becoming of a products performance and most people don't care to realise what bandwidth limiting this measurement actually means.
Stereophile's 80kHz measurements of the hypex vs frequency are most telling, because as you can see, the distortion rises very rapidly as frequency goes up. Hypex own measurements hint at this but you are not told the full story. This is what happens when the amount of feedback you can use is limited by stability margins and how an amplifier does this internally is particularly interesting. Clearly the hypex curtails its high frequency feedback very quickly.
TI, hats off to them, actually show you the difference between the two in their datasheet for their new TPA3251D2 class D single chip solution.
http://www.ti.com/lit/ds/symlink/tpa3251d2.pdf
You can see this here on page 10.
Figure one gives a graph that looks remarkably similar to hypex own datasheet, but figure 2 looks far more like stereophiles. Figure 2 is much more interesting.
Whose errors? Stereophile is in error or manufacturer's data is in error? From what 5th element said, it seems that Stereophile's measurements are closer to the truth as they did not bandwidth limit to 20kHz, thereby suppressing third order HD as was done by manufacturer's own measurement.
I haven't done anything more than filtering and minor changes to amplitude of test and reference tracks for clip2. Summation of left and right channels of Zoom recordings and level matching to mono right channel of reference for clip 2 seem to make comparisons to reference easier.
No DRC/convolution; this would require IR of drivers.
AP and AES point out how to reliably measure, what can and will go wrong, how to double check. Hypex datasheet follows that advice, Stereophile doesn't.
Theta isn't Hypex, so that does make things a little uncertain. Hypex did however limit "high end'' OEM Ncore buyers in how they are allowed to "improve" the Hypex product
There's nothing inherently wrong with how Stereophile measured the hypex amplifier. The output filter in hypex amps is very well designed and filters out a lot of the carrier residual. The only thing that increasing the measurement bandwidth would do is raise the noise floor and increase the absolute minimum that the distortion could reach as a result.
Stereophile's measurements concur actually very well with Hypex's own measurements given the higher noise floor encountered. The main difference in distortion performance is in the higher frequencies where Hypex ignores the harmonics by bandwidth limiting what they can measure and Stereophile did not.
Yes, you can change the filters and procedure used to eek out the last drop of performance for a given frequency (and Stereophile should really have done this for the 1kHz power vs distortion test). But like I said the interesting part of the amplifiers performance is between 5-20kHz. This is the part I want to see rendered more clearly, I don't care that the noise is higher, unless the noise is so high that I cannot see the distortion because of it - which clearly isn't the case here.
Besides you can clearly see that the main distortion component is 3rd harmonic as is illustrated in Stereophile's figure 10. The distortion residual is effectively clean from high frequency modulation and is very clearly resolvable. The AP is perfectly capable of processing the signal it has been presented, the only problem this is going to create is increasing the effective noise floor.
Stereophile's measurements concur actually very well with Hypex's own measurements given the higher noise floor encountered. The main difference in distortion performance is in the higher frequencies where Hypex ignores the harmonics by bandwidth limiting what they can measure and Stereophile did not.
Yes, you can change the filters and procedure used to eek out the last drop of performance for a given frequency (and Stereophile should really have done this for the 1kHz power vs distortion test). But like I said the interesting part of the amplifiers performance is between 5-20kHz. This is the part I want to see rendered more clearly, I don't care that the noise is higher, unless the noise is so high that I cannot see the distortion because of it - which clearly isn't the case here.
Besides you can clearly see that the main distortion component is 3rd harmonic as is illustrated in Stereophile's figure 10. The distortion residual is effectively clean from high frequency modulation and is very clearly resolvable. The AP is perfectly capable of processing the signal it has been presented, the only problem this is going to create is increasing the effective noise floor.
I think the "Elements" here refer to the classical earth, water, air, and fire, with the 5th being the quintessence from out of this world. The film by the same name is not referring to Boron, element with atomic number 5. Yes, a clever screen name nonetheless. Maybe we should call 5th element Leeloo?
Not to get too off-track, but I thought it worth looking at the statistical implications of these subjective blind tests.
Round 1:
sample of 85
drivers in test 7
expected value if randomly voted 12.14
standard deviation of response sample 7.27
Round 2:
sample of 50
drivers in test 8
expected value if randomly voted 6.25
standard deviation of response sample 4.02
In the first test only one value fell outside of a standard deviation from the mean expected value (speaker b with 25 votes). In the second test, two speakers fell outside of a standard deviation from the mean expected value (speaker e with 13 votes and speaker h with zero votes). These three results are probably significant (very statistical phrase, that). Most of the rest of the vote results in the two tests are really too close to the expected mean value to be of use with such a small sample size.
From test one, speaker b (a 'winner') is the Vifa TC9FD. From test 2, speaker e (a 'winner') is the ScanSpeak 10F/8424 and speaker h (a 'loser') is the PA130-8. It is really tough to make any conclusion with regards to the rest of the speakers tested based solely on the votes.
I don't know if anyone else cares, but at least the three above speakers would have been fairly unlikely to receive the votes that they did if the votes were totally randomly cast. This all assumes we have the same definition of what 'good' means with regards to speakers, that this 'goodness' falls within a normal distribution in the speaker population, that no one was peeking at votes, and that the testing itself was fair and unbiased.
Round 1:
sample of 85
drivers in test 7
expected value if randomly voted 12.14
standard deviation of response sample 7.27
Round 2:
sample of 50
drivers in test 8
expected value if randomly voted 6.25
standard deviation of response sample 4.02
In the first test only one value fell outside of a standard deviation from the mean expected value (speaker b with 25 votes). In the second test, two speakers fell outside of a standard deviation from the mean expected value (speaker e with 13 votes and speaker h with zero votes). These three results are probably significant (very statistical phrase, that). Most of the rest of the vote results in the two tests are really too close to the expected mean value to be of use with such a small sample size.
From test one, speaker b (a 'winner') is the Vifa TC9FD. From test 2, speaker e (a 'winner') is the ScanSpeak 10F/8424 and speaker h (a 'loser') is the PA130-8. It is really tough to make any conclusion with regards to the rest of the speakers tested based solely on the votes.
I don't know if anyone else cares, but at least the three above speakers would have been fairly unlikely to receive the votes that they did if the votes were totally randomly cast. This all assumes we have the same definition of what 'good' means with regards to speakers, that this 'goodness' falls within a normal distribution in the speaker population, that no one was peeking at votes, and that the testing itself was fair and unbiased.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.