LM3886 Parallel Use

[rljones]

ssanmor and Opie,

I'll be happy to send you my schematics and even by PCB layout once I've debugged it and make sure that all is OK. I use ExpressPCB (http://www.expresspcb.com/) for my small runs and I could email you the file and you could then order how ever many you want.

I powered one of the boards up this morning and found one tracing that needs changing, and I'll probably play with some other values through this week.

As for your schematic without the servos, I agree with paulb that the app note and the way I'm working up the amp is probably best done with the servos. This does require another +/- power supply for the 12 to 15V for the LF411CN opamps (I switched from 15 to 12 V because the AD8610 can only take up to +/- 13V). This isn't too difficult though, just tap off of the +/- 35V LM3886 supplies running through a couple of 3-legged regulators.

As for inversions, I'm not inverting anything in the amplifier, but I'm running it balanced. It you want it single-ended, just ground the negative inut.

[Nicke]

Another solution to drive it single-ended can be found at http://sound.westhost.com/project20.htm
It should work better than to just ground one input.

Niclas

[ssanmor]

Beware with that circuit applied to the LM3886!
What that circuit does is to use the non-inverting part as the master, its resistors set the overall gain. The output is then used as the input for the other stage, that is an inverting one with unity gain. LM3886 are not stable at unity gain.
To make it work, you can do two things:

a) Connect the output of the first stage to the second one after being attenuated. Then set the gain of the 2nd stage to the same amount. Then the amplifier has a gain (for example, 20) but the overall result is that its output has the same amplitude as the output of the 1st stage (but inverted).

b) Directly excite both stages with the same input, but one stage is connected inverting and the other non-inverting. See:

http://david.lewander.com/projects/amp/amp.htm

Suggestions or corrections?

[AudioFreak]

Originally posted by ssanmor
My idea is this (I will try to explain it as clear as I can):
I am going to use four identical non-inverting amplifiers like the one shown in this picture. The output R in parallel with L is substituted by a 0.1Ohm/5W resistor, (all picked by-hand to minimize differences between them).
Then two of them are excited by the direct input signal from my preamplifier (with a gain=1 buffer). This signal is inverted by an inverting opamp with gain=1 and then drives the other two power stages.
The outputs of the two top stages are connected together and to the + terminal of the speaker; The outputs of the bottom stages are also connected together and to the - terminal of the speaker.

All the gain resistors are also picked by hand to match them within 0.1%

As you can note, I don't use servos. The stages are AC coupled, so they don't amplify the DC offset from the preamplifiers.

Does this set-up have any chance of working well? If you don't think so sincerely, please send suggestions.

Thanks!

you cant replace the 10ohm resistor with a 0.1ohms resistor ....the 2 resistors perform 2 different task so there must be a seperate 0.1ohm resistor @ the output of each chip ... as per the schematic i posted right @ the beginning of this thread.

Originally posted by paulb

But it does not help with the DC offset from the amplifiers themselves. The 0.1 ohm resistors you're proposing may be too small to keep the two amps from "arguing" as Nelson says.
Didn't that App note from National warn about trying to do this without the servo?
If the two amps have, say, a 100mV difference in offset, this would be 0.5 amps through the balancing resistors (100 mV / (0.1+0.1))!

this is correct, for decent performance you should definately use the servos .... also, note that this does not guarantee that there wont be DC @ the output it just assures that all the chips in parallel create the same small DC offset....

Originally posted by Nicke
Another solution to drive it single-ended can be found at http://sound.westhost.com/project20.htm
It should work better than to just ground one input.

Niclas

Actually, while these amps are not unity stable, this may just work because of the clever way it is done but personally with chip amps like these, i wouldnt try to bend the rules just use the original method for bridging as per the schematic @ the start of this thread which was taken almost straight from the national application note.

Hope all this helps ....

AudioFreak.

[rljones]

the bridging technique described at http://sound.westhost.com/project20.htm is not quite true bridging; it is more of a hybrid. i believe it is more commonly referred to as a master-slave amp.

by having the output of the master amp driving the input to any number of slaves, all sorts of phase anomalies are created at the final output. i would avoid this design for high end audio (maybe its ok for a car stereo system where phasing is already chaotic, but not for a home system). this is what gave bridged amps a bad name several years ago.

the method originally described by AudioFreak at the start is better; my only suggestion was to use all non-inverted configurations and to keep the whole thing fully balanced with one 'set' for the positive input signal and another for the negative input signal. (i believe this balanced/bridged configuration, using the LM3886 devices, is essentially what is done in many of Jeff Rowland's products.)

i would also second (or third) the opinion that 0.1 (or maybe even 0.2) ohm resistors are essential for proper behavior of each output device, and that one resistor/inductor is also needed in most situations to prevent the amp from oscillating (and thereby overheating). This resistor/inductor should probably be used on each side of a bridged amp's load, but only on one side in a single ended version.

[ssanmor]

rljones,
I agree with you, what you say seems sensible.
But what do you mean when you say "using one set for the positive input and another for the negative input"? you mean providing adjustable gains for both phases of the signal in order to fully balance the waveform? Please make this a bit more clear.

And, do you think that the resistor + inductor in the speaker terminals would be mandatory for using a parallel-bridge setup and a 4 ohm woofer (this is my application ;-) ?

Thanks.

[rljones]

ssanmor,

it's not the impedance of a speaker that interacts with the amp. if all speakers were purely resistive loads, life would be easy. however, it's the inductance and capacitance of not only the speaker, but the crossover of the speaker and the cabling that's running to your speaker that affects the amp. unless you can see what's happening to your amp on a 'scope, i'd put in the resistor/inductor. many builders also include a 0.1 microfarad cap in series with a 10 ohm (or sometimes slightly smaller) resistor. this resistor-cap goes from the positive output lead to ground on a single ended amp. these go by the name Zobel or Boucherot networks.

as for the 'using one set' business, what i mean is that there is an input op-amp as a buffer (the lf411cn in audiofreak's first schematic), but running in non-inverted mode, driving a paralleled pair of lm3886 running in non-inverted mode. this is what i'm calling a 'set'. into this is put the + signal. another one of these 'sets' is used for the negative input. this is now a balanced amp. (you might want to search on the internet, or within this forum, balanced versus single ended amps.)

[rljones]

i'm going to continue on this same thread since it involves parallel lm3886s. i'll try to keep this as short as possible (and assume the above posts were read).

i finished my board and ran tests on it today. the results seem very good (described below). it also led me to incorporate some changes that will make it easier to run both single ended (read RCA input) and balanced (read XLR input), without having to change the output (bridged) strategy. once i'm sure the newer PCB is working well, i'll post the file for anyone who wants have it made at ExpressPCB (http://www.expresspcb.com/).

as mentioned in an earlier post, i'm using non-inverting inputs. the test circuit used a pair of opa627 (one each for the +/- input legs). each buffer amp drove a paralleled pair of lm3886. each of these pair had lf411cn servo opamps and had their output through 0.1 ohm resistor and then onto a resistor/inductor series element as described in previous posts, and then finally to the 4 or 8 ohm test load. power supply was +/- 35 VDC with 330 VA transformer. no input caps were used.

the following specs were found (all were at 1 W/8 ohms unless noted):

single ended input/single ended output (4 or 8 ohms to ground; ie, not bridged):
60 W into 8 ohms or 106 W into 4 ohms at 1kHz
freq resp: -0.2 dB at 10 Hz and -1.5 dB at 100 kHz (limits of test equip)
%THD: 0.012 at 20 Hz, 0.007 at 1 kHz, 0.017 at 20 kHz
noise (shorted input): 175 microvolts (80 microvolts with 400/80 kHz filter; note this is on test bench with open wires, aligator clips, and 2 adjacent computers)
DC offset = 2 mV no load; 4 mV at 1 W 20 kHz

balanced input with bridged output:
280 W no load
170 W at 8 ohms
230 W at 4 ohms
% THD: 0.009 at 20 Hz, 0.009 at 1 kHz, 0.030 at 20 kHz
freq respon and DC offset same as above
noise: 200 microvolts (same disclaimer as above)

single ended input with bridged output:
54 W into 8 ohms and oscillated (spike limiter kicked in) at 4 ohm load

these findings are very good. output would probabaly not droop as much with a larger transformer; i plan on using a 1 kVA per channel along with 2 of the above boards per channel (maybe 3, which would mean 12 lm3886 per channel). i did not listen to the amp. i will first make some changes described next, and then build a stereo unit for listening. one point i must mention is that in an earlier post i referred to transformer input. i tried two differen ones and was not able to adequately drive the paralleled lm3886s, so i instead turned to the buffered op amp method.

as for the main modification that i alluded to at the start of this post, it deals with the single ended input issue. as you can see from the data, a single ended input with a bridged output has lower power than when using the same input but with a single ended output (ie, not bridged). this is probably due to the two circuit halves 'bucking' one another. to eliminate this and to obtain maximum output using the whole bridged circuit, i'm going to use two separate inputs: one via RCA and one via XLR. the XLR version is completely balanced input to bridged output. this is the circuit i tested above.

for the RCA input, i'm going to add two more buffer op amps and then run them into the XLR circuitry. the two op amps will simply take the single ended input and one will forward it to the + input op amp (in the XLR input section); the other op amp will invert and pass it onto the - input op amp. the reason for using one inverting buffer is obvious: it gives the necessary out-of-phase signal for a balanced signal. the other op amp is simply used to treat the non-inverted signal as closely as possible to the inverted half. the other design change i'll make on the PCB is to allow use of either SOIC devices like the AD8610 as well as 8-in dip op amps like the opa627 i used in the prototype for the buffered input op amps.

one other area to test is to see if using a lower noise op amp instead of the lf411 would lower the noise floor. does anyone know if this would help in lowering noise?

[paulb]

Your servo amp has an ultralow bandwidth, like < 1 Hz, right? I can't see how a different op amp would make any audible difference. Noise at these frequencies is termed drift.

[rljones]

yes, the filter i believe is a 2nd order low pass at less than 0.5 Hz. i was wondering if the inherent noise of the op-amp accompanies the DC component and if this was potentially reducible by using another op amp.