New P482 Stereo Power Amplifier

[SmartOne_2000]

Thanks. Just had a look, it's been a long time since I looked at the Weiss.

Essentially, whilst it is a very good performing part, I see little in the way of a compelling argument to use it over the 1612. Quite similar in fact.

The biggest detractor however is the price. Why would you pay hundreds of dollars when you can get the same performance for less than $10?

Also, our buffer outperforms it. . As I have mentioned before, and as the Buckeye and Apollon amps demonstrate, a buffer based on the Purifi reference circuit using the 1612 is not high enough performance to take advantage of the new modules.

One other thing to mention is that to get the best performance out of these devices you need to pay a lot of attention to surrounding passive components and the PCB design and layout.

Just compare how "messy" the Buckeye noise floor is to our P482 or P801.

Sorry for the late reply ... my PC decided to crash on me and spent the entire day trying to fix it ... in the end, I had to use a network backup to get it running. Missed the past 2 weeks of work as a result.

Ok, now I'm curious to know in what way the Ultra-Buffer outperforms the Weiss amp. The amp has an extremely low noise floor (-180dB and lower) with the highest distortion components around -160dB (2nd harmonic). All these specs outperform the 1612 op-amp, I believe.

Regarding the potential usage of this amp in your future product, I was thinking of maybe offering it as a higher-priced option that one selects at checkout. Obviously current buffer board will have to be slightly modified to handle this option.

It's very true. Your Gen 2 amps indeed have very, very low noise floors, outperforming whatever has been measured out there.

Just a thought ... you might want to show the noise floor (inputs shorted) and standard 1kHz THD+N plots that you have on each amp's thread on your main page. This further shows confidence to users and competitors alike that your amps outperform anything out there.

 

[Alan March]

Our buffer is lower noise. Also possibly lower distortion at the relevant voltage and load range used (would need to implement a Weiss to be sure). That's not going to happen cos of the ridiculous price of the Weiss. We won't be having it as an option because we really don't want to get into this "up-selling" marketing nonsense that other manufacturers indulge in.

How many times has Amir done these uber super duper op amp swaps and found no difference?

The Weiss does not have a -180dB noise floor. Not physically possible. You also just can't look at these maximum values in their spec sheet. Ie at max voltage output. You need to look at relevant output voltages and loads.

Yep, might update the product pages with more info.

 

[Alan March]

Further to the above, the spec sheets need to be really carefully interpreted. The Weiss and OPA input noise density is pretty much the same. Our buffer achieves lower output noise than both.

Again, it's not just down to the op amp, but how the overall circuit performs. For example, Im not sure where the -180dB noise floor claim was derived from, but it is physically impossible

For example, assuming the 0dB reference is the maximum required buffer output of 10 volts, the intrinsic thermal noise of a 1k resistor (the most basic of circuit components) at 20deg C is much higher than -180dB.





So under perfect conditions, with a perfect amp you are simply not going to see noise levels or SINAD levels below (approximately) this level of -145dB.

In the real world you won't get close to it.

I suspect there may be some mis-reading of the graphs in the data sheet. Don't mistake an FFT noise floor for the actual total noise over a 20Hz to 20KHz bandwidth.

FFT noise floor is proportional to the resolution of the FFT. It's "FFT gain". I can explain that further if you would like

 

[SmartOne_2000]

View attachment 453
Thanks for the primer, Alan. I welcome the offer to explain the concept of FFT gain further, as you've suggested. Thanks!

The above plots were what I was alluding to, particularly the -190dB noise floor shown in the distortion residual FFT. I agree that this unusually low noise is most probably a result of a much larger FFT sample size or FFT resolution, as you indicated. But the harmonics are not reduced in the process, right?

The claimed THD+N specs from roughly -2.5dBu (-0.58 Vrms) to +22dBu (9.75 Vrms) are well under -130dB...which would further improve with higher load impedances (assuming increased thermal noise is not an issue). I've often wondered how this was achieved, assuming these are not Spice-based simulations but actual laboratory measurements. And if that's the case, I'm curious to know what instrument was used to measure their data.

Thanks again!

PS
BTW, I do not want to derail this thread with my Weiss posts. You may move them to an appropriate one if that's more appropriate.

 

[mikewxyz]

I received my P482 amp the first week in December and I was delighted. After listening to it for six weeks, long after the new audio equipment thrill has worn off, I’m still delighted. I use the low gain setting with a pre-amp volume dial positioned mostly at 9 o’clock. My speakers have a sensitivity around 90dB.

If you agree with the composer Claude DeBussy when he said music is the silence between the notes, then the P482 is very musical. It is quiet. I don’t hear any 60Hz hum or background noise. Female vocals with minimal accompaniment are a perfect showcase for this contrast.

The amplifier’s power easily handles loud complex orchestral passages with no congestion.

Fulfillment was easy. It took a week to get to N. America, the same time it usually takes to get across the country. Fit and finish quality is high.

Highly recommend.

 

[Alan March]

View attachment 453
Thanks for the primer, Alan. I welcome the offer to explain the concept of FFT gain further, as you've suggested. Thanks!

The above plots were what I was alluding to, particularly the -190dB noise floor shown in the distortion residual FFT. I agree that this unusually low noise is most probably a result of a much larger FFT sample size or FFT resolution, as you indicated. But the harmonics are not reduced in the process, right?

The claimed THD+N specs from roughly -2.5dBu (-0.58 Vrms) to +22dBu (9.75 Vrms) are well under -130dB...which would further improve with higher load impedances (assuming increased thermal noise is not an issue). I've often wondered how this was achieved, assuming these are not Spice-based simulations but actual laboratory measurements. And if that's the case, I'm curious to know what instrument was used to measure their data.

Thanks again!

PS
BTW, I do not want to derail this thread with my Weiss posts. You may move them to an appropriate one if that's more appropriate.

I will come back to the topic of FFT techniques in a later post but let's look at the graphs you have posted. They are not directly comparable, showing different things in different ways. The test conditions are different.

First thing is we must look at is the output voltage. The OPA is tested at 3 volts. The Weiss graph is measured in dBu. 3 volts is 11.7 dBu.

The Weiss thd+n at 11.7dBu is about -138dB.

The OPA (at 1kHz and 2k load - I will explain why in a moment) is about -137dB.

Why 2k load?. Simply that the Purifi modules input impedance is 2.2k.

I'm sure you will point to the rising distortion at higher frequencies of the OPA. Well this is erroneous in this context because this only happens at lower load impedances. There is very little effect at 2.2k load.

Secondly, the measurement is taken with 80kHz bandwidth. This leads to an erroneous results as it is taking into account all the harmonics above 20kHz which you simply can't hear and your speakers can't reproduce . You need to look over 20kHz bandwidth. Also with stimulus signals under 10kHz as again the second harmonic is 20kHz and all other harmonics above this so inaudible.

As an aside, our buffer has about 4 times the output current capability of the OPA, so doesn't get as affected at higher frequencies with lower loads.

So looking at those graphs the Weiss, in the intended application, is not really any better than the OPA.


Another graph. At 10 volts output you can see the OPA can be well below -140dB THD+N, even with 80kHz bandwidth.

 

[Alan March]

To re-iterate, the Weiss is a very good part. It is, certainly in this application, broadly similar in performance to the OP1612. However, it does have disadvantages.

It's form factor makes it more difficult to optimise overall circuit layout. It has very higher power consumption, note the heatsink. It uses 42mA sat there doing nothing. Price is prohibitive. 30 X the cost for no real benefit over the OPA1612. Finally it is not as good as our buffer

 

[SmartOne_2000]

I received my P482 amp the first week in December and I was delighted. After listening to it for six weeks, long after the new audio equipment thrill has worn off, I’m still delighted. I use the low gain setting with a pre-amp volume dial positioned mostly at 9 o’clock. My speakers have a sensitivity around 90dB.

If you agree with the composer Claude DeBussy when he said music is the silence between the notes, then the P482 is very musical. It is quiet. I don’t hear any 60Hz hum or background noise. Female vocals with minimal accompaniment are a perfect showcase for this contrast.

The amplifier’s power easily handles loud complex orchestral passages with no congestion.

Fulfillment was easy. It took a week to get to N. America, the same time it usually takes to get across the country. Fit and finish quality is high.

Highly recommend.

Thanks for the review, Mike ... can't wait for my P801 monoblocks sometime this year, I pray!

 

[SmartOne_2000]

I will come back to the topic of FFT techniques in a later post but let's look at the graphs you have posted. They are not directly comparable, showing different things in different ways. The test conditions are different.

First thing is we must look at is the output voltage. The OPA is tested at 3 volts. The Weiss graph is measured in dBu. 3 volts is 11.7 dBu.

The Weiss thd+n at 11.7dBu is about -138dB.

The OPA (at 1kHz and 2k load - I will explain why in a moment) is about -137dB.

Why 2k load?. Simply that the Purifi modules input impedance is 2.2k.

I'm sure you will point to the rising distortion at higher frequencies of the OPA. Well this is erroneous in this context because this only happens at lower load impedances. There is very little effect at 2.2k load.

Secondly, the measurement is taken with 80kHz bandwidth. This leads to an erroneous results as it is taking into account all the harmonics above 20kHz which you simply can't hear and your speakers can't reproduce . You need to look over 20kHz bandwidth. Also with stimulus signals under 10kHz as again the second harmonic is 20kHz and all other harmonics above this so inaudible.

As an aside, our buffer has about 4 times the output current capability of the OPA, so doesn't get as affected at higher frequencies with lower loads.

So looking at those graphs the Weiss, in the intended application, is not really any better than the OPA.


Another graph. At 10 volts output you can see the OPA can be well below -140dB THD+N, even with 80kHz bandwidth.

View attachment 461

Thanks for the analysis as always, Alan. It would seem the OPA was designed with the Gen2 Purifi modules in mind, with their 2K input impedances, to extract maximum performance from these modules. Obviously, one can't say for sure if this was ever a design factor for the OPA.

So, why would Weiss build a discrete opamp, knowing full well it doesn't offer any technical benefits over the 1612?

Also, curious to know if you measure the Purifi modules by themselves to see if they match their datasheet specs?
Also, do you know how we went from a -137db THD+N from the OPA to -115dB THD+N or so for the overall amp? That 20dB of loss is huge. Not saying -115dB is not an excellent figure.

BTW, for some reason, I can't seem to access the graphs I attached earlier. I get an "Oops, wrong page" type message. Are you able to access them?

 

[Alan March]

FFT processing. A simple explanation. (Please no one pick me up on pertinent details I have glossed over )



This plot of the Weiss on face value looks really impressive. However, there are a few things we don't know. Critical information that they haven't published with the graph.

First, we don't know what voltage the fundamental signal is at 1kHz. Note that its notched out so it doesn't affect the ADC of the measurement system. So, if it were 1v, it would be an easy load. 10 volts would be far more challenging and yield much higher harmonic distortion.

Secondly, the noise floor is not at -190dB. This is a product of the FFT processing. Often called FFT Gain. We do not know the FFT resolution. Without this we cannot calculate the actual total noise level.

When measuring the total noise level you need to specify the frequency range. Obviously for our purposes we are primarily interested in what we can hear, i.e 20Hz to 20kHz. Noise outside this range should not be ignored, it is of interest for a number of reasons when you are designing, but it won't be directly audible or for that matter even reproduced beyond about 30kHz as most speakers response really falls off around there.

One way to measure the total noise is with a simple AC volt meter. It would need to have a filter up front to limit it's frequency response from 20Hz to 20kHz. This could give you a direct number in uV RMS.

More typically these days you would use a measurement system that digitises the signal. Such a system can be used like an analogue volt meter and calculate the voltage from the acquired signal waveform. This is what Amir does for example with those power on plots. However, again he should state the measurement bandwidth and limit it to 20Hz to 20kHz. The number is the total noise in the specified frequency range.

Of course once digitised we can do all sorts of processing to the acquired signal waveform. Often performing FFT processing to look at the signal in the frequency domain.

In simplistic terms the FFT beaks the frequency range up into a number of individual blocks. These are often called bins or lines. The more bins, the higher the frequency resolution we have. If you have a frequency range of 20kHz and use an FFT resolution of 1000 lines, each bin would be 2Hz wide.

From this we can display what's happening in each bin, the contained signal and it's amplitude. So what if we look at a noise signal. Well noise can vary greatly in nature, but this purpose let's consider it it's random in but also have constant density across the measured frequency range.

Lets consider a noise level of 100uV across a 20kHz range. Let's apply a FFT at a really low resolution of 2 lines. This splits the frequency range into 2 sections, 0Hz to 10kHz and 10kHz to 20kHz.

The total noise is now split over those 2 bins. As such it's power level in each bin drops by half or 3dB. If the FFT resolution was 4 lines, the noise in each bin would drop by 6dB. So you can see that the higher the FFT resolution, the lower the indicated noise floor will look. There is no actual limit to the number of lines you can have, however it takes progressively longer to process.

This is great as it allows you to see what's going on below the real total noise floor, to look at very low distortion signals and spuria.

There is a formula you can use to figure out the FFT gain. A representation below showing the RMS total noise level and the lower FFT noise floor.



So regarding the Weiss total noise floor, we know that it is not -190dB. Thats just FFT processing gain in action. From other information in their documentation we know it's more like -140dB ref 3volts.

 

[Alan March]

Thanks for the analysis as always, Alan. It would seem the OPA was designed with the Gen2 Purifi modules in mind, with their 2K input impedances, to extract maximum performance from these modules. Obviously, one can't say for sure if this was ever a design factor for the OPA.

So, why would Weiss build a discrete opamp, knowing full well it doesn't offer any technical benefits over the 1612?

Also, curious to know if you measure the Purifi modules by themselves to see if they match their datasheet specs?
Also, do you know how we went from a -137db THD+N from the OPA to -115dB THD+N or so for the overall amp? That 20dB of loss is huge. Not saying -115dB is not an excellent figure.

BTW, for some reason, I can't seem to access the graphs I attached earlier. I get an "Oops, wrong page" type message. Are you able to access them?

I don't think the OPA was designed with any knowledge of the Purifi modules. It has been around since 2009. As I alluded to earlier, it is getting close to the limits possible and physics starts to become a limitation. The noise created by electrons jiggling around in a resistance due to heat

I would not say the Weiss has no advantages over the OPA1612. If you did a full detailed analysis you may conclude it just nudges the OPA into second place. However, it does depend on the application, the surrounding design etc etc. Remember that any real world application will not achieve the results in the data sheets.

Why would Weiss design such a part? Well I'm sure they will give a number of reasons why they think it's better, some may have creedence, some will be just marketing. However, the main reason will be to make money selling the part!

We both know how audiophiles can convinced to part with money .

Luckily the Weiss op amp really does live up to the hype in terms of performance. It really is very good. It's appalling value for money, but I have no doubt it does what is says on the tin.

Unfortunately most, maybe all of the other "boutique discrete" op amps out there don't live up to the hype and won't come close to an OPA1612.

 

[Alan March]

Also, do you know how we went from a -137db THD+N from the OPA to -115dB THD+N or so for the overall amp? That 20dB of loss is huge. Not saying -115dB is not an excellent figure.

The overall performance is of the buffer and power amp module. It's a complex summation.

The power amp module will amplify the noise and distortion of the buffer. Remember there is about an additional 12dB of gain. It will add its own noise and distortion. That worsening of THD+N is quite normal.

Again, you have to be extremely careful when interpreting and comparing these numbers.

That -115 thd+n for the overall amp is at a specific output condition: 5 watts into 4 ohms which is 4.475 volts RMS. So whilst we saw the opa1612 earlier providing -137 dB thd+n, that was at 3 volts output.

This is not the output voltage of the buffer when the overall amp is outputting 4.475 volts. We have to subtract the gain of the Purifi modules of 12.3dB. That means the buffer is outputting only 1 volt. This lower output voltage means lower dynamic range and lower THD+N because at this output level noise is the dominant component of the THD+N figure.

If we use the OPA1612 as a reference, from the graph below it's thd+n is about -126dB at 1 volt. So for he whole amp, after going through the Purifi modules, you would lose about another 13dB in overall THD+N, down to about 113dB.

Our buffer performs better than the OPA1612, hence we see up to about -116dB (-115 in the specs to account for component performance variations).

 

[SmartOne_2000]

Thanks, Alan! ... This is an excellent analysis, tutorial, and well-written summation of audio measurements we all deal with. Thanks again. I was not aware that the noise floor in a 1 kHz or 10kHz spectral plot was simply the FFT gain. From your explanation, it appears that there's no limit to this noise floor, as it's a function of the number of FFT bins and processing time. Good to know! (thanks!).

You may want to pin these tutorials so that anyone searching your site for information about your products won't have to dig through all the posts on the forum. And, in my opinion, if I read good technical tutorials on a website, I view that as an indication of the technical prowess of the company, garnering more confidence in their products.

BTW, some audio vendors now seem to prefer to use the OPA1656 (see attached) over the 1612. It does seem to have some excellent specs. Do you know why, and are they justified in doing so? I know it's newer than the 1612, so there must be a reason why Texas Instruments decided to make a new low-noise, low-distortion opamp for audio applications.

And for a head twister ... You said "Our buffer performs better than the OPA1612, hence we see up to about -116dB (-115 in the specs to account for component performance variations)." ... so the ultra buffer board composed of 1612s performs better than a 1612 ?

Thank you again for taking the time and effort to educate us on audio specifications and to properly read a datasheet when comparing one product against another.

 

[Alan March]

BTW, some audio vendors now seem to prefer to use the OPA1656 (see attached) over the 1612. It does seem to have some excellent specs. Do you know why, and are they justified in doing so? I know it's newer than the 1612, so there must be a reason why Texas Instruments decided to make a new low-noise, low-distortion opamp for audio applications.

I do know why and no they are not justified.

Again this is a choice made by people that don't have a good understanding of electronics. It's obvious that they didn't perform any testing of the 1656.

The reason why they are using it is due to a misunderstanding of something Bruno Putzeys said in an interview. An electronics DIYer and audio blogger asked Bruno about alternative better op amps. You could see from Brunos pausing that he didnt really have a a good option in mind, but under a bit of pressure from the interviewer he suggested that the 1656 could perform better in some applications and might be worth a try.

This has been misinterpreted by some as a formal recommendation for a better option. It ain't. Anyone who has actually tested it knows that is the case.

The 1656 is FET input opamp with extremely high input impedance. It would indeed work better than the 1612 in certain applications, but not this one. It has higher noise, especially at lower frequencies. Plainly obvious in the Buckeye and Apollon test data.

 

[Alan March]

And for a head twister ... You said "Our buffer performs better than the OPA1612, hence we see up to about -116dB (-115 in the specs to account for component performance variations)." ... so the ultra buffer board composed of 1612s performs better than a 1612 ?

Is it composed of 1612s? There are other topologies beyond the Purifi reference circuit used by most.

 

[Oniiz86]

@Alan March In the recent Apollon 1ET6525SA amp review, a member had mentioned "another thing to point out in favor of Apollon's buffer is its very healthy (stated) 100 kOhm input impedance. This kind of immunity is rare these days.", I see that most Purifi OEM assemblers opt for a higher input impedance of either 94 kOhm or 100 kOhm, is this much higher input impedance compared to the 20 kOhm of your amps simply chosen because of their use of the OPA1656 op amp or is it simply for ease of driving source components more comfortably but I recall you mentioning that 20 kOhm is more than sufficient?

 

[Alan March]

@Alan March In the recent Apollon 1ET6525SA amp review, a member had mentioned "another thing to point out in favor of Apollon's buffer is its very healthy (stated) 100 kOhm input impedance. This kind of immunity is rare these days.", I see that most Purifi OEM assemblers opt for a higher input impedance of either 94 kOhm or 100 kOhm, is this much higher input impedance compared to the 20 kOhm of your amps simply chosen because of their use of the OPA1656 op amp or is it simply for ease of driving source components more comfortably but I recall you mentioning that 20 kOhm is more than sufficient?

100k input impedance is only beneficial for a small number of badly designed tube pre amps whose Hi-Z low current outputs can't pull the skin off a rice pudding.

Totally unnecessary for any well designed tube pre amp (we have many customers using tube pre-amps) and pretty much any solid state source.

The high input impedance also cause issues with noise and DC offset. It's not a function of the 1656, it's set by input resistors.

 

[SmartOne_2000]

Is it composed of 1612s? There are other topologies beyond the Purifi reference circuit used by most.

No doubt ... does this topology also prevent the sudden rise in distortion at high frequencies as is observed in all class D amps tested so far on ASR, as seen below? If so, it would be a first and make an excellent and compelling selling point to potential customers.

 

[SmartOne_2000]

@Alan March, should time permit, would you be willing to generate the above sweeps for your Gen2 amps (p801 or p482)? Simply looking at the profiles and not the absolute values of the THD+N spec, since you said your source is quiet but still noisier than desired.

Secondly, how about the power curve below? I looked at the stated 100W THDN spec of -122dBf or the p801, and it is so darn close to the module spec itself (maybe -123dB?). This means your buffer hardly interferes with the module's performance, and its noise levels are well under -130dB at this data point, correct?

 

[Alan March]

No doubt ... does this topology also prevent the sudden rise in distortion at high frequencies as is observed in all class D amps tested so far on ASR, as seen below? If so, it would be a first and make an excellent and compelling selling point to potential customers.

View attachment 470

Not this again .

There is *NO* problem with high frequency distortion in class D amps.

Again it's dumb ass Amir and others not understanding what they are doing or looking at.

If you apply a 15kHz signal as seen in the graph you posted, the 2nd harmonic is at 30kHz. The 3rd harmonic is at 45kHz.

YOU CANNOT HEAR THESE HARMONICS
YOUR SPEAKERS PROBABLY WONT REPRODUCE THEM
THEY ARE IRRELEVANT TO SUBJECTIVE QUALITY

Measurement bandwidth should be 20kHz because that's what you can hear .*Not* 45 kHz. If it were the case you would not see any rise in 10 or 15kHz distortion.

If you want to test the audible effects of distortion from high frequency signals perform an intermodulation test with 19 and 20kHz tones.