Not sure what you mean, measure in-room frequency response with the same speaker using different amps, after some sort of basic level matching?
If so: No, I don't. That's way too indirect a measurement to me. I do measure amps with varying (resistive) loads, that's required to get full power specs anyway so I measure frequency response while I'm at it.
And as I said, unless we're talking Class-D with post-feedback output filter there's nothing to see, a fraction of a dB at most, normally very much less.
Attached find a plot from a rather cheap small commercial Class-D of the 'bad' kind which requires a certain resistive load to obtain flat response in the treble.
One can see both the influence of less or more than optimum high frequency damping (Q) of the 2nd-order LC filter as well as the influence of DC coil resistance (giving about 0.4dB wideband difference) which limits output resistance.
Attachments
I have 3 of the Ralph Morrison books, all from about 1990. I recently purchased them used for 10 to 15 dollars each.
But the Ralph Morrison books are:
a] small, the biggest has 228 pages.
b] some of the chapters are about complying with 1990 US NEC code.
c] they are written using field theory rather than circuit theory (this will be a challenge for some)
The most useful one may be:
"Grounding and Shielding in Facilities"
I'd rather have a book with 200 pages of content, as is the case with Morrison's book, than a book with 200 pages of content buried in 300 pages of fluff, as is often the case with textbooks.
In my opinion, it's an advantage that Morrison's book starts with the EM field theory. If you don't understand how the fields work and interact, you won't understand how EMI becomes a problem in the first place, how shielding works, etc. His "intro to field theory" chapter is pretty accessible, actually. If you were awake in high school physics class, you should be able to follow that with a minimum of material refresh.
I like that he focuses on developing understanding rather than memorization.
I'll check out the Grounding and Shielding in Facilities. As you point out, the older editions are often available at a substantial discount. I employ that trick as well...
~Tom
In my opinion, it's an advantage that Morrison's book starts with the EM field theory. If you don't understand how the fields work and interact, you won't understand how EMI becomes a problem in the first place, how shielding works, etc. His "intro to field theory" chapter is pretty accessible, actually. If you were awake in high school physics class, you should be able to follow that with a minimum of material refresh.
I like that he focuses on developing understanding rather than memorization.
I'll check out the Grounding and Shielding in Facilities. As you point out, the older editions are often available at a substantial discount. I employ that trick as well...
~Tom
Sounds like you'd find a spectrum analyzer rather useful. I will caution tools capable of even sort of measuring the error spectrum of a well executed composite amplifier are expensive (refer to the earlier mentions in this thread about reducing loop gain to make errors big enough to measure) and require care in use. The audio interfaces I use with an RTA for this purpose max out at 110dB DnR. That's barely better than the Modulus-86, meaning one has to rely on averaging out noise and careful setup to ensure one doesn't end up just looking at noise picked up by the measurement gear.
In regards to supplies I'm unsure where to find the conventional engineering wisdom you mention. Certainly, there's abundant evidence supplies are audible and, if you search over the power supplies forum, you find threads discussing about how 120 to 140dB PSRR is desirable to maintain an amp's SnR against ripple in the supply's reservoir capacitors. Optimized discrete amps rarely exceed 90dB at their best frequencies and rather performant control amplifiers which normally operate from unregulated supplies---notably the LME49811---are still down to 80dB at 20kHz. So one wants to get about 50dB more PSRR from somewhere, whether it's an audio SMPS (small and potentially inexpensive but kind of hard to get), linear supply regulation (big pass devices with heatsinks and regulator design complexity on both rails), or additional loop gain from the control device in a composite amp to suppress power stage errors (very compact albeit with design complexity of its own).
Tom calls out high PSRR as a key feature of the Modulus-86. I very much agree with this. Since this is a block diagram thread I'll refrain from implementation details. But note that the LME49710 the M86 uses as a control device is exceptional in its own right---used optimally it pushes 100dB SnR at 20kHz.
Should you choose to check the assumption a resistive load is a sufficient approximation of a driver and its associated air volume please do share the results. I'll be quite interested.
Powerful tools would be good. The direction I'm heading is comparative analysis of time domain captures, the type of thing that the program Diffmaker nominally does - unfortunately the latter is almost unusable, far too flawed; so, I working, very slowly, on creating my own version of such.
Engineering "wisdom" in regard to supplies has come over the years mainly from my own experiments in effectively making the rails as stiff as possible, to as high a frequency as possible. In PSRR numbers I would be looking for around at least 100dB everywhere that could be relevant to the circuit, anything less would start to concern me, and going for significantly more I would see as rather pointless. I've used optimised banks of smoothing caps, carefully thought out regulation, local decoupling, shaping of the waveforms that occur on the leads of the transformer - all ideas have proved beneficial, in terms of the subjective qualities in the sound.
Engineering "wisdom" in regard to supplies has come over the years mainly from my own experiments in effectively making the rails as stiff as possible, to as high a frequency as possible. In PSRR numbers I would be looking for around at least 100dB everywhere that could be relevant to the circuit, anything less would start to concern me, and going for significantly more I would see as rather pointless. I've used optimised banks of smoothing caps, carefully thought out regulation, local decoupling, shaping of the waveforms that occur on the leads of the transformer - all ideas have proved beneficial, in terms of the subjective qualities in the sound.
Yeah... You pretty much need an AP SYS-2700 series or the new APx555 to get better than 110 dB dynamic range in a distortion analyzer.
You can also see my measurements. My Modulus-86 measures the same high performance both on a well-regulated lab supply and on a real supply (toroid + rectifier + caps), whereas, the LM3886 THD degrades by nearly an order of magnitude when used on a real supply. There may be some confounding variables, but the most likely cause of this is that the Modulus-86 offers much, much greater PSRR than the native LM3886.
An amplifier with loop gain. It's a wonderful thing.
~Tom
Tom, will you putting up any harmonic distortion measurements?
I think I remember reading where someone suggested a way to correlate measurements to audio impressions. It was waaaaaaay out there but kind of made sense. I thought it was in the "Sound Quality vs. Measurements" thread but I couldn't find it there. Did anyone else see this besides me?How do you measure an amps ability to image? I don't think this has much to do with a flat FR.
Last edited:
See Post #1. I need to update these as the LM3886 wasn't making good thermal contact with the heat sink (I was using a SilPad 400 rather than thermal compound). That caused the IC to trip the thermal limiter before it reached max output power. I'll post new results here as well as on my Modulus-86 page once available.
See the links posted by twest in Post #79.
~Tom
Neither post #1 or the link to your product page show the higher order harmonic distortion profiles. I’d like to see the 2nd through the 11th @100, 1K, 10K Hz at full power and at 1W. The Sean Olive link was not the discussion I was referring to, I remember that discussion thread was somewhere on diyaudio.
Good luck with that; a straight line of slope -sqrt(power) on a THD+N plot means you're seeing noise floor, not harmonics.
I am a little curious as to the cause of little kink at -106dB THD+N at 16W, though it looks like it might be a change in range selection within the measurement gear to me---wouldn't be the first time I've seen a composite amp hand the test equipment its own limitations back. I'm more curious as to why you're interested in harmonics to 110kHz at these aren't audible; planning to use the amp for a non-audio purpose?
Do they ever get to hard data or testable hypotheses? I stopped looking after the first couple pages of "I'll neither reason about or share the data I have", "I'll allude to but avoid saying how I think it works", and "I'd rather speculate than do ABX testing" type responses. Don't get me wrong; not saying folks shouldn't push hard problems around to try to understand and solve them, just saying I wouldn't mind skipping over most of the pushing.
Yeah, the board needs a visit to a lab with more resolution capability
The complete absence of higher order harmonics can make an amp sound sterile and uninvolving. I'm not saying this design is but getting a look at the upper harmonic profiles might show us the root character of the amp even though it is a very deep root indeed.
It seems like each of these modules are self contained and paralleleing/bridging them is convenient, does that mean there is no requirement for the ballast resistors described in the National App Notes should someone want to put more amps out? I know dumping 11A/channel should be enough for any god forsaken loudspeaker but when the chip is asked to do this, I can't imagine it doing it effortlessly. This ties into a question about using a 47940 instead. I'll assume Tom probably looked at this part for a parallel chip implementation and decided against it for the layout complications it would introduce. He explains the parasitic inductance on the upper bandwidth that is so vital to the outer loop of this circuit so he is right when he says the layout is the circuit. What about it Tom, what's the headache with the 49740?
In the photos Tom provided you can see there is a location for a ballast resistor on the board right before the speaker output terminal to make paralleling feasible.
What's the point of using a quad OpAmp when only single one is needed per section? Even for a dedicated board design as a BPA (4 sections) it doesn't make much sense, both from a layout perspective as well as from feasibility. Duals would be OK for paralleled sections, though.
This project is about an universal and versatile amplifier module, not about cost-optimized production designs. If you want to save some cents a single THAT input section on one board might be used for driving multiple parallel power stages but that depends on the exact circuit Tom is using there. Likewise, one might share some regulators etc...
Last edited:
I've been fooled by straight lines like that several times when I didn't check slope carefully. Some amps do have monotonically increasing distortion with decreasing level which looks like a noise slope but actually is very annoying low level distortion.
Those kinks are in fact from the range switching on the AP once a DUT is good enough.... but sometimes can come from the DUT as well. Using a pre-divider at the analyzer input reveals what is coming from DUT and what is not.
Well, at $60/board, bridge and paralleling is not just a few cents. Is there any reason an LM1875 or LM3875 couldn't be enveloped in the same way by a precision op-amp?
That is indeed the case here.
The little kink in the THD vs power curve at 16 W is indeed caused by a range change in the AP SYS-2712. That's what happens when your amp pushes the limits of even state of the art test gear...
Yeah, the board needs a visit to a lab with more resolution capability
Until last month when Audio Precision released the APx555, the SYS-2700 series I used for my testing was the state of the art distortion analyzer. It is capable of measuring THD+N down to -115 dB. The APx555 pushes this to -120 dB.
Unless you find a lab that managed to get their paws on one of the new APx555s, you won't find a lab with better equipment than what I used. If you are aware of equipment that beats the performance of the Audio Precision gear, please list the make and model number, as I would be most interested.
That's funny, because the tube crowd (in particular the single-ended triode (SET) crowd) says the exact opposite. My own subjective experience has been that I prefer a flat THD+N vs frequency profile as amps with this THD profile sound more open and natural to me. Both my DG300B and Modulus-86 deliver a flat THD+N vs frequency profile. If you want flat and low THD, the Modulus-86 is where it's at.
As an aside, a stereo Modulus-86 with chassis, power supply, heat sinks, and all can be built for about 75 % of the cost of the Electra-Print output transformers I used in my DG300B. Just saying... The MOD86 doesn't glow as nicely in the dark, however...
As KSTR has pointed out, the ballast resistors are included on the board. To connect two Modulus-86 boards in parallel, you connect the inputs in parallel and the outputs in parallel. That's it. Bridge and bridge+parallel are just as easy. No component substitutions required.
The LM3886 data sheet states 11.5 A typical; 7 A minimum. That's part of the spec table, hence, tested in production. If the device cannot supply 7 A of output current, it is tossed in the scrap bin. The 11.5 A is found by lab characterization. I don't know specifically about the LM3886, but usually a statistically significant number of devices are tested and the distribution of the performance parameter is found. From that, the number for the spec sheet table is generated. That's how the industry works.
The devices don't have feelings. They don't break a sweat or go on strike if you ask them to work hard. 11.5 A out of a sizable BJT output transistor sounds perfectly reasonable. Note, however, that this is tested using a short current pulse. It tells you something about the instantaneous current sink/source capability, which is relevant for the transient response.
I think you answered your own question, actually...
For a single channel, only one LME49710 is needed, so there's no reason to use a quad part. Most builders will likely build a single-channel setup, so addressing that market seemed reasonable. For those who wish to bridge/parallel/bridge-parallel, they can connect multiple boards together easily. I included the bits needed for this on the board, so all you have to do is to connect the boards according to the schematics in the design doc, and you're in business. This makes it possible to build a 4-channel amp and turn it into a 2-channel bridged (or paralleled) amp or a mono block bridge-parallel amp at a later date if desired.Using a quad opamp in a bridge-parallel board would turn the most sensitive part of the circuit into a rat's nest. It would result in considerably worse performance than using four single op-amps. I chose to optimize for performance rather than trying to shave a few cents off the build cost.
~Tom
Last edited: