Hello there,
I tried to find info comparing different LM4780 amps out there, but found that info is scattered. I thought it might be a good idea to gather them together in one place.
I am a happy user of PD version of LM3886 gainclone. I would say, gainclone is the circuit that uses minimalist's approach. I started building LM 4780 kit from PD, but someone suggested me parallel 86. And then I noticed there are so many different versions.
PA03 aka Pavel Dudek's 4780 amp
Parallel 86 by Tom Christiansen
Sympatico by Twisted Pear Audio
Audio Sector Minimalist Approach
ETC
There are also commercial 3886 amps such as:
AKITIKA GT-101
AmpsLab Synergy
Audio Sector
ChipAmps
ETC
What are the main difference between them? Have you actually listened to them? What's your listening impression?
Thanks,
Doug
I tried to find info comparing different LM4780 amps out there, but found that info is scattered. I thought it might be a good idea to gather them together in one place.
I am a happy user of PD version of LM3886 gainclone. I would say, gainclone is the circuit that uses minimalist's approach. I started building LM 4780 kit from PD, but someone suggested me parallel 86. And then I noticed there are so many different versions.
PA03 aka Pavel Dudek's 4780 amp
Parallel 86 by Tom Christiansen
Sympatico by Twisted Pear Audio
Audio Sector Minimalist Approach
ETC
There are also commercial 3886 amps such as:
AKITIKA GT-101
AmpsLab Synergy
Audio Sector
ChipAmps
ETC
What are the main difference between them? Have you actually listened to them? What's your listening impression?
Thanks,
Doug
The Parallel-86 uses two channels of the LM4780 in parallel for higher output current. The Parallel-86 can deliver minimum 14 A peak of output current. This is enough to drive a 4 Ω load all the way to the ±35 V supply rails. If you lower the supply voltage to ±28 V, the Parallel-86 will drive a 2 Ω load with vanishingly low THD.
The Parallel-86 is a composite amplifier. In the composite topology, a precision op-amp is used to perform error correction on a power amplifier. The end result is low noise and very low THD. The THD of the Parallel-86 is almost 10 dB below the measurement capabilities of my Audio Precision APx525 analyzer.
The Sympatico is a composite amplifier as well. It uses the two channels of an LM4780 in series for a bridged amplifier. Each channel of an LM4780 is guaranteed to be able to deliver up to 7 A of output current, but with the amplifier channels in series, that means the maximum output voltage across a 4 Ω load is 4*7 = 28 V.
This means a bridged amplifier based on the LM3886 (or LM4780) will not be able to deliver any more power than a single LM3886 (or half LM4780). To me, that defeats the purpose of the bridged design.
In addition to the incredibly low THD, the composite amplifier topology also results in incredibly high power-supply rejection ratio. I measure the same performance of the Parallel-86 regardless of if I'm using a regular non-regulated linear supply, a pair of $1200/each well-regulated lab supplies, or a switching supply (Connex SMPS300RE). With non-composite chip amps, you can count on a degradation in THD+N of about 15-20 dB (5-10x) over the performance specified in the data sheet (see attached measurement).
One note on the LM4780: TI has decided to discontinue the LM4780 along with a handful of other ICs. The LM4780 is getting hard to find in low quantity. I picked up a tube of LM4780s and offer them for sale with my Parallel-86 boards while supplies last.
Tom
The Parallel-86 is a composite amplifier. In the composite topology, a precision op-amp is used to perform error correction on a power amplifier. The end result is low noise and very low THD. The THD of the Parallel-86 is almost 10 dB below the measurement capabilities of my Audio Precision APx525 analyzer.
The Sympatico is a composite amplifier as well. It uses the two channels of an LM4780 in series for a bridged amplifier. Each channel of an LM4780 is guaranteed to be able to deliver up to 7 A of output current, but with the amplifier channels in series, that means the maximum output voltage across a 4 Ω load is 4*7 = 28 V.
This means a bridged amplifier based on the LM3886 (or LM4780) will not be able to deliver any more power than a single LM3886 (or half LM4780). To me, that defeats the purpose of the bridged design.
In addition to the incredibly low THD, the composite amplifier topology also results in incredibly high power-supply rejection ratio. I measure the same performance of the Parallel-86 regardless of if I'm using a regular non-regulated linear supply, a pair of $1200/each well-regulated lab supplies, or a switching supply (Connex SMPS300RE). With non-composite chip amps, you can count on a degradation in THD+N of about 15-20 dB (5-10x) over the performance specified in the data sheet (see attached measurement).
One note on the LM4780: TI has decided to discontinue the LM4780 along with a handful of other ICs. The LM4780 is getting hard to find in low quantity. I picked up a tube of LM4780s and offer them for sale with my Parallel-86 boards while supplies last.
Tom
Attachments
Whilst I haven't listened to those particular incarnations of the LM3886 I would suggest (along with tomchr it seems) that PSRR is probably a fairly good measurement to use to get to grips with their differences.
If you're going to build a good PSRR LM3886 amp and not go for a composite you'll probably go the way Akitika went and have only a single, regulated supply. I'd suggest avoiding any designs which don't address the PSRR weakness of the bare naked LM3886.
Tom,
Thanks for the detail info. It is very helpful.
With LM4780 soon going to Dodo, and because you also have LM3886 board, which do you recommend to drive Magnepan MMG (4ohm)? I don't listen to the music loud. Mostly I play them as background while working on others, so moderate power will be OK.
Doug
My current LM3886 (Peter Daniel version) is dead quiet with the recommended minimal power supply. I even used this amp for >100dB horn, and it was dead quiet.
But I guess it actually has poor THD?
With the recommended power supply (Power-86 + Antek AS-2222 transformer), the Modulus-86 can deliver 40+ W into 8 Ω and 65+ W into 4 Ω both with extremely low THD.
The Parallel-86 will deliver 60 W into 8 Ω and 120 W into 4 Ω using the recommended supply (±35 V). You can get about 225 W into 2 Ω, but you need to lower the supply voltage to ±28 V.
The Modulus-86 does offer the lowest THD on the market, so unless you really need the extra power, I'd go with the Modulus-86.
"Dead quiet" would refer to the noise floor or residual mains hum. That's what you hear with the music off, but the amplifier on.
What I measured in the graph in Post #2 was the impact of the power supply on the amplifier THD when the amplifier is delivering a signal to the load. I suspect the reason many chip amplifiers sound kinda strained and harsh is due to the interaction between the supply and the amplifier. The composite topology of the Modulus-86 and Parallel-86 remove this supply/amp interaction to the point where it disappears below the measurement noise floor. The result is an amplifier that sounds open, natural, and clear-as-a-bell transparent.
Tom
I would guess by 'dead quiet' you mean that you hear little or no hiss when there's no music playing. This isn't a good way to evaluate the PSRR because with nothing playing the power supply doesn't have to work very much at all, so there's very little ripple.
THD is measured with a sinewave as stimulus which is quite unlike music. So THD in and of itself isn't particularly meaningful to perceived audio quality.
However, having said that most audio measuring devices don't measure THD, rather THD+N and that difference is crucial with an amp which has poor power supply rejection - the power supply noise at the output of the amp does contribute to the 'N' part of this measurement. Which is what you see in the plot tomchr included - the difference between an amp with a clean PSU (regulated, low noise) and a normal PSU (just a bridge and res caps).
The THD isn't the only parameter I measure. I do agree that measuring only THD isn't particularly meaningful. However, the complete picture presented by measurements of THD, output noise, output power, THD+N vs frequency, THD+N vs output power, IMD, etc. is meaningful if you know how to interpret the results.
Actually, no. The reason for the difference in THD+N of an LM3886 when using a regulated supply and unregulated supply (shown in Post #2) is because the load presented by the amplifier on the power supply creates signal-dependent variations in the supply voltage. This shows up as THD (not noise) on the output of the amp. It's the THD part of the THD+N that rises.
A good amplifier will have high PSRR, hence, reject these signal-dependent variations in the supply voltage. As a result, it will show better THD and THD+N performance at high output power.
I tend to use THD as a synonym for THD+N and count on the reader to have sufficient context to understand my point correctly. What's shown in the plot in Post #2 is THD+N (measured using an Audio Precision SYS-2700 - two measurements combined in MS Excel).
Tom
Last edited:
Do you have FFTs which show that its purely harmonic distortion? I'd be interested to see them if you do.
I don't have proper measurement equipment, and I don't have reference level speakers and source, most of all, my home is more like children's theme park rather than listening room, so I have no means of doing high fidelity reproduction of sound at home. So I mostly am very happy when my built amplifiers are quiet when the input is turned off.
That's why I only mentioned about dead quietness about my 3886 amp. When paired with Harbeth Compact 7 ES3, it produces somewhat bloated mid-bass, more like baritone male voice area. I don't know if this is typical for 3886 amp, or if it's just me or my system, or if it can be fixed with more advanced (complicated?) designs.
That's why I only mentioned about dead quietness about my 3886 amp. When paired with Harbeth Compact 7 ES3, it produces somewhat bloated mid-bass, more like baritone male voice area. I don't know if this is typical for 3886 amp, or if it's just me or my system, or if it can be fixed with more advanced (complicated?) designs.
It doesn't look like I saved the FFTs. Next time I play with the LM3886 by itself, I'll make sure to save the plots.
I'm curious how you arrive at the conclusion that signal-dependent ripple on the supply would result in a rise in the overall noise floor.
Tom
All my amps will meet this criterion.
Without measurements, I can't determine what you mean by "bloated mid-bass". The amplitude response of the Modulus-86 and Parallel-86 is flat vs frequency.
Tom
you hear it bloaty?
fine, it may not be the amp at all, as you described your listening environment is not suitable for high fidelity. the environmnet contributes a lot, the least problematic is to have a decent freq.response from the amplifier(s). the next problem is the speakers, and then the enivronment is the most defining one.
i think you need to get an eq.
surely eq is from the devil and all that stuff but heck. if it sounds better with it, then it does.
fine, it may not be the amp at all, as you described your listening environment is not suitable for high fidelity. the environmnet contributes a lot, the least problematic is to have a decent freq.response from the amplifier(s). the next problem is the speakers, and then the enivronment is the most defining one.
i think you need to get an eq.
surely eq is from the devil and all that stuff but heck. if it sounds better with it, then it does.
I look forward to seeing them.
It could simply be that your meaning for noise is rather narrower than mine. To me 'noise' isn't just thermal noise. The non-harmonic components of power supply ripple to me count as 'noise' as they're hardly harmonic distortion - which has a very precise definition.
I'm curious how you arrive at the conclusion that the signal dependent power line frequency-related ripple on the supply would count as 'harmonic distortion'.
Last edited:
Probably what you're hearing is the power supply as the simple LM3886 lacks sufficient PSRR to count as transparent. You could try adding more capacitors to your supply see if the 'bloat' you report changes in tonality.
That's entirely possible. To me, noise is random processes. White noise, for example, is random both in amplitude and frequency. It is uniformly distributed in frequency and normally distributed/gaussian in amplitude. Thermal noise is white. As is shot noise (the noise caused by current through a semiconductor junction). 1/f noise is normally distributed in amplitude but slopes down with increasing frequency.
I actually didn't say anything about the mains frequency or mains ripple. You are correct that in an amp with poor PSRR, the mains frequency (and harmonics of it) are likely to mix with the desired signal and form IMD products. That's intermodulation distortion (IMD), not noise, and is decidedly not random.
What I was referring to is the supply ripple caused by the desired signal, specifically by the load current. Ignore all mains hum for a second. The power supply has a finite and non-zero output impedance. Hence, any supply current will cause the supply voltage to droop. Hence, if you have the amp deliver, say, 1 A peak to the load at 1 kHz, you will see a supply current of 1 A peak (plus the idle current of the amp). The supply current in a class AB amp will be half-sine shaped pulses with a frequency of 1 kHz. The VCC rail will contain the current from the positive output swing. The VEE rail, the negative. This current is not pretty and due to the finite non-zero output impedance, it will set up a series of harmonics of the signal frequency (1 kHz) on the supply voltage. It will cause a signal-dependent ripple voltage at the signal frequency (1 kHz) and harmonics thereof (n*1 kHz). Some of this ripple/droop will make it through the LM3886 due to the low PSRR and cause an added signal at harmonics of the signal frequency. So you'll have signal (1 kHz) + harmonics (n*1 kHz) of the signal. That's harmonic distortion.
The reason the regulated supply results in lower THD with an amplifier with poor PSRR, is that the output impedance of the regulated supply is much lower, hence, less droop occurs when the amp starts delivering current to the load.
Tom
Last edited:
Why would it merely be 'entirely possible' ? If you know in your own mind what you mean by noise then ISTM you'd know if your own meaning was different from mine. Or did you feel my meaning was insufficiently clear?
Do you disagree that in the current context the analyser couldn't care less about whether the noise originated from 'random processes' or not?
Again 'random' or 'not random' is a red herring here ISTM. So given the current context how could the analyser tell the difference between intermodulation distortion and noise? From my understanding they're both going to contribute to the residual which is the output of the notch filter. It's that residual which is plotted in the graph you showed is it not?
<snip discussion about harmonic distortion on the rails, its presence isn't disputed>
I ignored mains for a second while I read your text but seems that second is a very long one - you've not returned to it. Are you going to ignore it forever?
Slap some typical numbers down on these handwaving terms ('much lower', 'poor PSRR') and we can have a discussion and hopefully at least one of us will learn something.
Last edited:
The mains frequency is either ~50Hz or ~60Hz.
But there are an infinite number of other random frequency pulses contained in the mains signal.
I don't believe these random frequencies give rise to ONLY harmonic distortion.
In my view most of this interference, that gets through the amplifier, is "noise".
But there are an infinite number of other random frequency pulses contained in the mains signal.
I don't believe these random frequencies give rise to ONLY harmonic distortion.
In my view most of this interference, that gets through the amplifier, is "noise".
If you look at the FFT of the output signal of an amp when running on a regulated supply and when on an unregulated supply, there's no difference in the noise floor. On the unregulated supply, there are some mains-related components starting at 120 Hz (assuming 60 Hz mains frequency) and up to about 1-2 kHz. I suppose these could be the "random frequencies" that you mention, but I've always found them at integer multiples of the mains frequency.
I'm sure there's a bunch of other crap on the mains. Loads switching on and off. Switching supplies with broken EMI filters, etc. I've never seen any of that cause a rise in the noise floor or frequency components on non-integer multiples of the mains frequency present on the amp output, though. Perhaps they've been filtered out by the reservoir/smoothing caps of the power supply.
If you watch the FFT of the output signal as the amp delivers more and more power to the load, you'll see the THD creep up. That's the case with any amplifier on any supply, assuming you can get the harmonics to show above the noise floor.
With an unregulated supply, on a typical LM3886 circuit, you will have some mains related components from the rectification present on the amplifier output. As the load current increases, the rectification/mains related hash will increase in amplitude as well. So you see an increase in mains-related harmonics on the amp output. This gives rise to an increase in IMD (n*60 Hz + desired signal). So the desired signal and harmonic thereof (due to THD) will have a skirt of mains-related hash around it (due to IMD). All this IMD is deterministic. It occurs at combinations of n1*Fmains + n2*Fsignal. The analyzer will interpret this as noise, hence report a degradation in THD+N, unless an FFT or continuous sweep analysis is performed to extract the THD (no +N).
In addition to the mains ripple (and other rectification hash), there'll be a ripple component that's signal dependent. If you've ever watched the mains ripple on an oscilloscope when the amp is delivering significant power (say 25-30 W @ 1 kHz) to the load, you'll notice that you get the usual rounded sawtooth 120 Hz ripple with a little 1 kHz component riding on top. That little 1 kHz ripple will make it through the LM3886 supply pins as well. The 1 kHz ripple component is not a high-purity sine wave (say <0.0001 % THD) and the path through the supply pins is not clean. Also note that the PSRR on the VCC is much higher than on the VEE pin of an LM3886, so the signal-dependent ripple will cause a rise in THD.
One could tease out how much of the THD+N degradation is due to the supply impedance and how much is due to the mains ripple by adding resistance in series with the supply leads on the regulated supply. I hope to touch on some of this next time I update my Taming the LM3886 pages.
Tom
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.