• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Buckeye Purifi Eigentakt 1ET9040BA monoblock power amplifier Stereophile Measurements

Status
Not open for further replies.
the 4.7R case is of course not supporting my hypothesis as you say.

The 1% THD case has the amp in clipping for part of the time and here the feedback loop is powerless. this may raise the noise floor but that’s still much much below the distortion from clipping. Bruno’s loop design secures very clean and fast recovery from clipping both from voltage and current limit. This is an area that could be good to include in amp reviews.

Note that I work primarily on our transducers and not directly with the amps (I leave that to Bruno and Søren who are much better than me). As I understand, The faulty pilot batch had a misplaced protection diode that was causing trouble for one of the op amps. unfortunately not caught soon enough since it only showed up close to clipping and at high frequencies. A very frustrating mistake indeed.
 
the 4.7R case is of course not supporting my hypothesis as you say.

The 1% THD case has the amp in clipping for part of the time and here the feedback loop is powerless. this may raise the noise floor but that’s still much much below the distortion from clipping. Bruno’s loop design secures very clean and fast recovery from clipping both from voltage and current limit. This is an area that could be good to include in amp reviews.

Note that I work primarily on our transducers and not directly with the amps (I leave that to Bruno and Søren who are much better than me). As I understand, The faulty pilot batch had a misplaced protection diode that was causing trouble for one of the op amps. unfortunately not caught soon enough since it only showed up close to clipping and at high frequencies. A very frustrating mistake indeed.
The opportunist me cannot help myself:
Please,please make nice PSUs to go with your amps :)
 
Note that I work primarily on our transducers and not directly with the amps (I leave that to Bruno and Søren who are much better than me). As I understand, The faulty pilot batch had a misplaced protection diode that was causing trouble for one of the op amps. unfortunately not caught soon enough since it only showed up close to clipping and at high frequencies. A very frustrating mistake indeed.

Are you going to release the serial numbers of the defective batch?
 
the 4.7R case is of course not supporting my hypothesis as you say.

The 1% THD case has the amp in clipping for part of the time and here the feedback loop is powerless. this may raise the noise floor but that’s still much much below the distortion from clipping. Bruno’s loop design secures very clean and fast recovery from clipping both from voltage and current limit. This is an area that could be good to include in amp reviews.

Note that I work primarily on our transducers and not directly with the amps (I leave that to Bruno and Søren who are much better than me). As I understand, The faulty pilot batch had a misplaced protection diode that was causing trouble for one of the op amps. unfortunately not caught soon enough since it only showed up close to clipping and at high frequencies. A very frustrating mistake indeed.
Thank you for your sincere reply, Lars. This is much appreciated. And thank you for the explanation of the HF distortion issue, it really makes sense. And I agree that Bruno is a genius with respect to multi-loop feedback applied in class D amplifiers. And yes, recovery from clipping is fast and quite smooth, especially for class D amplifier, however I am not sure that the reviewers will be measuring it. It might be nice to do so in every amplifier review.

NC252MP_swing.png
 
And yes, recovery from clipping is fast and quite smooth, especially for class D amplifier, however I am not sure that the reviewers will be measuring it. It might be nice to do so in every amplifier review.
Clipping behavior is certainly worth investigating as it makes a real difference especially for smaller amps which are pushed into clipping more often (typically way more often than users might think).
And again we have the problem that fast and well-controlled soft-clipping (so that any feedback loops never go fully open) would be best in terms of audible impact but the game for highest 1% power number kind of prevents it. Say, with a soft-clipping reaching 1% at 90% output voltage the 1% power value drops to 0.81x of what could have been achieved with hard clipping, and with the soft-clipping starting at 80% for an even softer knee we'd be down to 0.64x which is deemed unacceptable despite of sounding much more listenable when driven into clipping.
 
Clipping behavior is certainly worth investigating as it makes a real difference especially for smaller amps which are pushed into clipping more often (typically way more often than users might think).
And again we have the problem that fast and well-controlled soft-clipping (so that any feedback loops never go fully open) would be best in terms of audible impact but the game for highest 1% power number kind of prevents it. Say, with a soft-clipping reaching 1% at 90% output voltage the 1% power value drops to 0.81x of what could have been achieved with hard clipping, and with the soft-clipping starting at 80% for an even softer knee we'd be down to 0.64x which is deemed unacceptable despite of sounding much more listenable when driven into clipping.
But again - a "normally" designed class AB amplifier, without evident design flaws, has far better clipping and recovery behaviour than any class D amplifier, including Ncore and Purifi. And it is free of that wild rise of noise floor above the clipping point. Below class AB clipping into 8ohm at 20Hz and 1kHz, for THD10% and THD1%.

THD 10% approx., 20Hz and 1kHz
A250W_8R_overload_20Hz_THD10%.png A250W_8R_overload_1k_THD10%.png

THD 1% approx., 20Hz and 1kHz
A250W_8R_overload_20Hz_THD1%.png A250W_8R_overload_1k_THD1%.png

I am no expert in class D amplifier design, but I am still curious if the clipping recovery and noise floor rise could be done even better than we have seen by now.
 
@pma, your last set of plots for AB amp doesn't seem to show better clipping than the previous one which of course may one of the better class-D designs wrt clipping.
Both are hard-clipping and show some rail-sticking and will not sound nice when clipping sets in. The trade off is maximised clean power.
As for class-D, ICEpower A(S)C series was also very OK (soft edges, minimal sticking) unless very severly overdriven last time I tested (like 10 years ago).

Voltage true soft-clipping is best done in front of the amp core (so that it never actually clips) but it is not trivial with unregulated rails. True soft-clipping of output current, obviously, is much harder to implement. Current-drive amps are easier to softclip both in voltage and current, for that matter.
 
clipping and class

There is no fundamental difference between class AB and D wrt to clipping. the clipping recovery is mainly determined by the feedback loop: The feedback wants to compensate for the lacking signal amplitude during clip which gives the ‘overhang’ and possible ringing when going out of clipping. The challenge is to tell the feedback loop to ignore the past and start from a fresh when going out of clipping.

and here class AB tend to have simple 1st order loops with high unity gain frequency where as class D needs lower unity gain frequency due to the sampling nature induced by the switching frequency. This means that class D needs a higher order feedback loop to achieve enough loop gain in the audio band. A high order loop is of course more challenging to design, stabilise and control wrt to clipping recovery. The methodology and modelling tools needs has been developed since the end of the previous century with substantial contributions from my own work (eg the discrete time modelling framework) and later Bruno’s brilliant work on extending the model to self oscillating loops and any duty cycle (my model was restricted to the idle condition only). This allows optimisation of loops with staggering loop gain that are stable to nearly 100% modulation index and recovers out of clipping.

Another complication of class D over AB is that the pulses need a minimum duration. this means we cannot continuously vary the duty cycle all the way to 100%. we can only have the minimum pulse or skip it. Adding to this we have that all self oscillating loops tend to lower the switching frequency as we get to full modulation. All of this is handled by our modelling framework and optimised in Matlab even before soldering.

Note that use of higher order loops can be applied to class AB and it is a lot easier task since it can be modelled in the continuos time domain. Performance of such amp can be insane and best the best class D. The only reason this is not seen widely is that class AB designers are typically not familiar with high order control theory. Bruno and I come from a past in Sigma-Delta modulated ADC and DACs where the high order loops is a necessity.

The only but huge advantage of class D is it’s high power efficiency and compactness. Everything else is a disadvantage including the difficulty of designing it.

Rise of the noise floor during clip:
1% THD at clip means we have an error at -40dB. I don’t think that the noise normally at -130dB rising a bit and mainly above 20kHz is of any concern.
 
The only but huge advantage of class D is it’s high power efficiency and compactness.
I would add "cost per high performance watt."
 
Voltage true soft-clipping is best done in front of the amp core (so that it never actually clips) but it is not trivial with unregulated rails. True soft-clipping of output current, obviously, is much harder to implement.

When I put my QSC amps under load and scope and drive them into their clip protection circuitry,
I see what looks like compression of the sine wave and not a flat-top truncation.
I've assumed, not being an amp man, that this is accomplished via compression circuitry at the line level before the gain stages kick in. Am I on the right track?

If I am, why wouldn't amps in general incorporate such 'compression limiting', instead of actually letting amps clip?

What I'm reading from you, Lars, and others, it looks like designing amps to clip gracefully is a bitch.
I guess I'm asking why bother with that, if there is an easier integrated line-level solution.
thx
 
As I understand, The faulty pilot batch had a misplaced protection diode that was causing trouble for one of the op amps. unfortunately not caught soon enough since it only showed up close to clipping and at high frequencies. A very frustrating mistake indeed.
@Lars Risbo

Would you kindly release the serial numbers of the defective batch?
 
@Lars Risbo

Would you kindly release the serial numbers of the defective batch?
I don't think it works this way. Manufacturers would inform the companies who bought defective modules who in turn would inform the buyers of their amplifiers and offer to replace them. We're not talking about an ecoli outbreak in carrots. I see no reason for a public dump of serial numbers of amplifier modules that are only available to OEM's.
 
When I put my QSC amps under load and scope and drive them into their clip protection circuitry,
I see what looks like compression of the sine wave and not a flat-top truncation.
I've assumed, not being an amp man, that this is accomplished via compression circuitry at the line level before the gain stages kick in. Am I on the right track?

If I am, why wouldn't amps in general incorporate such 'compression limiting', instead of actually letting amps clip?
PA amps like QSC in fact often implement a gain reduction loop, using an internal clipping detector and reduce gain until the clipping detector stops triggering.
So initially the amp clips for a few cycles but then tries to keep things safe. The gain cell is usually good enough for PA work in terms of distortion but in the current SINAD race it would keep the amp from reaching stellar specs, unless a very elaborate design is used.

And as for the alternative, static pre-clipping (at the input), the main reason why it is not widely used either is that it always reduces maximum undistorted (say, <1%THD) power measurements and nobody wants that, making you look worse than anybody else. And this pre-clipping, while conceptually simple, is actually a design challange if one wants to fully maximise undistorted output levels. Sagging rails need to be factored in in a practical way, and at different output current the hard-clipping level of the amp core (which is what we want to avoid) may be different as well, even with fully regulated supplies. That depends on the actual design of the amp core.

Ideally one would combine both methods, implement a soft-clipper to keep any transient clipping as benign as possible and then let a limiter kick in to keep the amp from constanly distorting. This is implemented is some studio monitors.
 
I don't think it works this way. Manufacturers would inform the companies who bought defective modules who in turn would inform the buyers of their amplifiers and offer to replace them. We're not talking about an ecoli outbreak in carrots. I see no reason for a public dump of serial numbers of amplifier modules that are only available to OEM's.

Your assumption that these are sold only to OEM's is false. These modules are sold directly by Purifi to end users on the Purifi site:


And yes, it would be nice to simply check with a manufacturer but many would prefer to "trust but verify".

Another reason to publish the serial number range is for those who buy these amps used- not all manufacturers will necessarily support these when they are no longer under warranty. Not all manufacturers may be around in a few years when someone wishes to verify the modules.

Purifi has always prided themselves on transparency. They have already stood up and taken responsibility for the release of a faulty "pilot batch". It would seem the right thing to do to protect customers going forward to take the next step and release what would seem to be a small serial number range to protect both their customers and their reputation. This is suppose to be their flagship product and now there is a defective group loose in the market. This taints the brand and causes doubt in the marketplace. The obvious and easy solution is to simply release the serial number range so that the defective modules can be identified and consumers can assure themselves that they are getting what they paid for. It is the obvious solution to the issue from both a business and moral perspective. Not doing so simply amplifies the doubt (pardon the pun) and casts a cloud over the brand. Since they have already admitted there was an issue, publishing the serial numbers would do nothing to worsen the issue but rather, would really be the only solution that would make the issue go away by allowing customers to resolve the doubt.
 
Last edited:
Batch Release testing is normal and accepted QC process. 100% Function testing and Batch testing for specific parameters is the usual practice. Although, also it is recommended to do 100% testing in Launch/Early release stage [e.g. first 100/1000 pcs or 90 days of serial production] until the manufacturing process stabilizes.

And it is normal to test only YOUR part of the manufacturing chain - in other words you trust your supplier to do proper testing himself.

But it looks that Purifi has done the right thing - recall and exchange. BTW I am also one of the early adopters of 9040 modules. so aleeady sent request to Apollon Audio to check whether my amps are affected too.

Have you got response from Apollon? Do you have masurments of your PET1200? Could you share some opinion on it?
 
regarding serial numbers and the recall of faulty units:

As we stated earlier: it was a pilot run from June that subsequently was recalled which means we notified the recipients (OEM early adopters). One of these modules somehow managed to escape the recall and ended up being reviewed. It is very unlikely that there are any more modules escaping the recall and even less likely that they ended in the hands of end users. We are of course double checking our logs to verify and will act accordingly but it takes a bit more time due to the holidays.

If anyone owning a 9040 still wants reassurance they are very welcome to query us with their specific serial numbers and we will check after the holidays.

Again, a recall means that we contact the owners directly and this recall was some time ago. Volume production has been running since then and the end of line test tightened.

The only reason we did a public statement was due to the review of a likely faulty module which could cast doubt about the modules general performance or the end product integration. Again, a recall means that we contact our customers directly.

Happy New Year to everyone!
 
regarding serial numbers and the recall of faulty units:

As we stated earlier: it was a pilot run from June that subsequently was recalled which means we notified the recipients (OEM early adopters). One of these modules somehow managed to escape the recall and ended up being reviewed. It is very unlikely that there are any more modules escaping the recall and even less likely that they ended in the hands of end users. We are of course double checking our logs to verify and will act accordingly but it takes a bit more time due to the holidays.

If anyone owning a 9040 still wants reassurance they are very welcome to query us with their specific serial numbers and we will check after the holidays.

Again, a recall means that we contact the owners directly and this recall was some time ago. Volume production has been running since then and the end of line test tightened.

The only reason we did a public statement was due to the review of a likely faulty module which could cast doubt about the modules general performance or the end product integration. Again, a recall means that we contact our customers directly.

Happy New Year to everyone!
I want to take this opportunity to make it publicly known (if there was any doubt) that I am in no way going to let Purifi take the blame/criticism all alone that some might throw at them.

Ultimately it was my build with one of these modules that slipped through, to a high visibility reviewer no less. Trust me when I say this has had me feeling thoroughly embarassed, not even so much at the outcome (as mistakes happen and I have always believed they are opportunities to learn and improve) but more so at the issue I have caused Purifi. I don't mind handling criticism (constructive or otherwise) but always feel bad when it affects or draws others into the fray.
 
Ideally one would combine both methods, implement a soft-clipper to keep any transient clipping as benign as possible and then let a limiter kick in to keep the amp from constanly distorting. This is implemented is some studio monitors.
Thx KSTR !
 
What I'm reading from you, Lars, and others, it looks like designing amps to clip gracefully is a bitch.
I guess I'm asking why bother with that, if there is an easier integrated line-level solution.
thx
yes, it’s difficult to clip gracefully but we have solved the problem at minimal cost. This also solves other issues that otherwise would need a separate solution do not much saving in leaving it off.

The line level processing is typically added to a system eg in the form of as limiters and compressors. Could also be a simple soft clip as you say. An active speaker may have advanced DSP limiters that beyond preventing hard clipping is also limiting driver excursion. All that is system dependent so not something to add inside the generic amp module.

If the amp module did not clip gracefully to the full rail and is soft clipped at the line level input then the module would need higher supply voltages to reach a given power level (soft clipped conservatively). this would drive up the cost of the system. the cost scales with the max volts times max amps levels.
 
I would add "cost per high performance watt."
yes and this pretty much follows directly from the higher power efficiency. This allows smaller power transistors to be used (no secondary break down issues etc), smaller power transformers and heat sinks etc. However, class D hinges on having an output inductor (without that it would be more like class A in efficiency) which needs to handle the power. this is a cost adder compared to AB
 
Status
Not open for further replies.
Back
Top Bottom