amplifier's power supply topology

[mimikos]

Hi all,

I am about to build a new stereo power amplifier with two integrated circuits. I know that it is better to use two transformers with two individual rectifier circuits, but i would like to avoid the cost of a second transformer. I am wondering if I will get better stereophonic figure by using one transformer and two seperated rectifier circuits in order to have seperated grounds for the two individual channels, instead of using one transformer and only one rectifier circuit for both channels. I mean, will the difference be detectable?

[wintermute]

I went down this path with my LM3886, I think it is a good middle ground. Provided that you have an appropriately sized transformer and a reasonable amount of capacitance in each channels separate rectifier/cap bank then I think that the result should be better stereo separation. (based on the premise that the modulation of the power supply by each channel will be isolated to their own individual cap banks), whether this premise stands up or not I'm not sure.

I did not however have separate grounds choosing to have the star ground common to both channels (but all returns separate).

As I did not build a single supply I can't make comments as to whether there's any audible difference (RMAA measurements should easily show if there is a measureable one).

I am going to rebuild the power supply for my other amp (which does have a single rectifier and cap bank) and converting to a rectifier and cap bank for each channel, effectively doubling the capacitance and will be taking measurements before and after, however with the speed I get around to things I wouldn't hold your breath waiting for the results

Someone else who has actually tried both, listened and measured might chime in

Tony.

[sangram]

I haven't measured it, but when I moved from a single rectifier to a separate one for each channel I got a very noticeable improvement in sound quality, and not just the imaging. I also ended up halving the cap bank from 80,000uF for both channels to 40,000uF to each channel because I didn't add capacitance, and did not lose much of the bass.

I went down to 20,000uF per channel later without further loss of impact, whereas with a single rectifier and 40,00uF (my first attempt) the sound was still a little lifeless and would grind down at higher output levels.

[abraxalito]

Originally Posted by mimikos

I am about to build a new stereo power amplifier with two integrated circuits. I know that it is better to use two transformers with two individual rectifier circuits, but i would like to avoid the cost of a second transformer.

I'm not so certain that it is so much better to use two transformers. One big transformer will have better regulation than two small ones. Also will be cheaper, alternatively the same money can be spent to get a more powerful supply.

I am wondering if I will get better stereophonic figure by using one transformer and two seperated rectifier circuits in order to have seperated grounds for the two individual channels, instead of using one transformer and only one rectifier circuit for both channels. I mean, will the difference be detectable?

I don't think crosstalk between channels is really a serious issue when running both left and right from the same power supply. Neither does Doug Self in his power amp book. Its true that grounding might be simpler with two separated supplies, but this issue is really very minor in my experience. I'm running a stereo pair of chip amps at the moment very happily from a single supply and haven't encountered grounding problems. I have compared this single supply system with a similar set-up using separate supplies and the differences are inconsequential.

[twest820]

Originally Posted by abraxalito

I don't think crosstalk between channels is really a serious issue when running both left and right from the same power supply.

Depends on your design targets. Consider the LM3886 as an example as it's one of the better chipamps around. Avol at 20kHz is 50dB so if its operated in a typical configuration with 30dB gain there's only 20dB excess loop gain available for error attenuation. If you assume an unregulated common supply for two channels with 20,000uF supply ripple will be about 12dB below the signal level. Vee PSRR at 20kHz is about 48dB. So, if the left and right signal levels have a constant power spectral density and are within 10 dB or so of each other (which in my experience is typically the case), then the channel separation is 48dB + 12dB + 20dB - 10dB = 70dB, or about 0.03% IMD. However, 1/f is a considerably better approximation of music's power spectral density so, if you start analyzing some slice of midrange or tweeter bandwidth as an IMD victim from the rest of the spectrum as lowpass filtered by the supply, that margin typically reduces by 20 to 40dB. 0.3% IMD has been shown to be audible and undesirable, 3% is pretty poor.

One can always run different analysis and come up with somewhat different numbers, but the finding getting a clean trebel is harder you might think has been a consistent one in my experience. The LME49810 makes better numbers in this regard than any monolithic chipamp I know. If you start looking at biamping or triamping with the tweeter channel as a victim operating the control loop on op amps like the LME49990 or LME49724 ends up being attractive for their higher Avol over the LME49810.

Originally Posted by mimikos

I am wondering if I will get better stereophonic figure by using one transformer and two seperated rectifier circuits in order to have seperated grounds for the two individual channels, instead of using one transformer and only one rectifier circuit for both channels. I mean, will the difference be detectable?

Work through the math above for your design options and then you'll know. The advantage of separate rectifiers is the ripple in the victim channel tends to be more spectrally compact so the IMD floor ends up being lower. However, every time I've worked through this I've ended up reaching the conclusion the extra USD 15 or so it costs to buy two smaller trafos and an extra bridge looked like cheap insurance.

Another advantage of paired trafos is you can mount them with opposing windings and get some amount of leakage flux cancellation; chipamp circuits tend to be compact enough they're not too susceptible to noise pickup from the trafos but I've seen amps hit surprisingly bad noise problems due to cap charging transients being being reradiated from transformers and picked up within input stages or the feedback loop. Shielding low frequency magnetics is hard, so getting a few extra dB of attenuation from field cancellation is a boon.

[sangram]

getting a clean trebel is harder you might think has been a consistent one in my experience

I concur. One cure (IMHO) is good decoupling, I usually use 100nF SMD ceramics right at the chip pins - P2P, not PCB so they're actually *on* the pin - and a 100uF Panasonic FM at the PSU entry points. Even so, the treble is not as 'fast' and clean as say my Pass Labs F5, which has even less PSRR (but a more beefy power supply). In the case of an AC-coupled chipamp, the capacitors at the input and Ci position play a pretty important role IME. I find very good results with a Panasonic FM at the Ci position, far better than the exotic Black Gate NX and Nichicon Muse that I've also tried there, specially for clean treble (unfortunately it is still a little behind a good discrete though far ahead of the cheaper Class D amps and most commercial amps under $1000).

In my case, if using a center-tapped transformer where stereo grounding becomes a real nightmare with just one transformer and two rectifiers. Often the best option is to relocate the star ground to the location where the power supply lead enters the chipamp, decouple it really well with low-ESR caps, and keep the ground paths really short, with a very thick ground wire back to the PSU caps. I still think dual mono is the better configuration.

About flux fields - I've used a LM4766 an inch away from a EI-core transformer without any hum being picked up at all. Maybe just lucky, I guess, or maybe not sensitive enough hearing

I don't think crosstalk between channels is really a serious issue when running both left and right from the same power supply.

Agreed. I think however there is more to dual supplies than crosstalk. I hope to get some test equipment soon, then I can have some measurements between dual mono and dual rectifier, but IME the dual mono just sounds a lot better!

[abraxalito]

Originally Posted by twest820

Consider the LM3886 as an example as it's one of the better chipamps around. Avol at 20kHz is 50dB so if its operated in a typical configuration with 30dB gain there's only 20dB excess loop gain available for error attenuation.

So far, so good, I'm following your argument.

If you assume an unregulated common supply for two channels with 20,000uF supply ripple will be about 12dB below the signal level.

But here I start to get lost. For a 50W/8R amp, our peak signal level is 56V p-p (20VRMS). 12dB below this is about *0.25 so 14V p-p. I've never seen such high ripple on the power supply in any amp I've designed. A 14V drop in 8mS from a 10,000uF capacitor (remembering I = C * dV/dt) implies a current draw of 17.5A whereas the peak current for 50W/8R is 28/8 = 3.5A per channel, 7A total maximum. So somewhere these numbers aren't stacking up. Secondly, the ripple will be at low frequencies, smallish multiples of the mains, and will fall with increasing frequency because the res cap's impedance is falling with frequency too.

Vee PSRR at 20kHz is about 48dB. So, if the left and right signal levels have a constant power spectral density and are within 10 dB or so of each other (which in my experience is typically the case), then the channel separation is 48dB + 12dB + 20dB - 10dB = 70dB, or about 0.03% IMD.

This part about IMD I don't follow either. How does PSRR translate into IMD? You're saying what's on the power supplies will intermodulate with the signal, rather than just adding to it?

One can always run different analysis and come up with somewhat different numbers, but the finding getting a clean trebel is harder you might think has been a consistent one in my experience.

I agree that a clean treble doesn't come so easily with chipamps, but my experience of that is that its caused by poor grounding and decoupling, not by interchannel crosstalk. I did have lots of treble grain on my active speakers (single channel per PSU - see my blog) but I fixed this by re-grounding.

[sangram]

abraxalito - are you sure about the 56V P-P figure for a chipamp running off say, 34V rails?

[abraxalito]

yes - that's peak to peak. with 34V rails, we have both -34V and +34V so if the amp could swing all the way to the rails we'd get 68V p-p. But it can't, so in practice with 34V rails we get around 62V p-p. 56V p-p is just a conservative figure because in practice we get a little more than 50W into 8R on bursts.

[sangram]

Right, I know in theory that's what it should be but I rarely see anything above 14V before the sound starts to crack up real bad. Mostly 11-12V, though most of real listening its 7-8V. I assume these are RMS figures as they come off a DMM. I'm just not sure it can actually manage 20VRMS output cleanly.