Neurochrome LM3886DR Build

Strange.. Everyone kept telling me not to go over 84 volt or else the chip would die. I have many of these little LM3886T, so I took the risk in testing one of them. I have a power supply that can produce +-36ct or +-48ct at 4amp. I am using a large size heatsink without a fan. I first tested with +-36 parallel of 6 speakers, probably some loss through the wires but this brought it down to 2ohm. It sounds great playing "AC/DC Who made Who". Pretty loud with lots of punch. The heatsink was warm but nothing hot. I then switched it to the +-48, this is 96v total, thinking it will die or not produce any sound because of the SPIKE. Well, it played the same "Who made Who" louder with stronger punches. I left it playing for days, thinking it could only max out for a short amount of time. It didn't give up. The heatsink was hot, but wasn't melting crayon hot. I measured the voltage at the chip's pins, taking in 96volt. when I turned it up full power, the voltage dropped a bit, but it was still 94volt.

Dumb question, but has anyone tried higher voltages? Could this be one of the defectives? I got these LM3886 directly from National Semi in the early 2000 as free samples. And later bought them from various people who claimed they were left over genuine parts for another productions. And all the parts seemed to be able to handle the higher voltage pretty well.
 
I wasn't there, but based on experience the designer(s) of the LM3886 were most likely given an initial set of specifications that the product needed to meet. During design and development there was likely a back and forth between designers and marketing on what the chip was actually capable of. If they settled on a maximum voltage of +/- 84V it's because the very competent designers engineered the chip to be able to exceed that specification. It's referred to as the designed in safety factor. All competent engineers do it this way.

Exceeding the published limits puts one in the safety factor area. Perhaps the chance of destroying $50 worth of electronics is not an issue (assuming one doesn't burn down a house during the event). If you're crossing a bridge over a deep canyon, it would be unwise to test where the safety factor of that bridge design ends.
 
±42 V (= 84 V total) is the absolute maximum supply voltage with input signal applied. ±47 V (= 94 V total) without input signal applied.

Now this does not mean that the chip will explode if you exceed the supply voltage by 1 mV, but you will impact reliability and you will over voltage the devices within the chip.

Also note that the LM3886 is only guaranteed to be able to deliver 7.0 A under all operating conditions. Thus, the maximum supply voltage suitable for 4 Ω operation is ±28 V.

Tom
 
Hello TioFrancotirador and Tomchr,
There is a very very easy way to get rid of the feedback cap and have an offset voltage of almost zero Volt. In the two LM3886 amplifiers I build, one amplifier has an offset voltage of 0,2 mV and the other of 0,9 mV. These offset voltages are very stable under unstable circumstances like varying output power and changing temperatures.
See: LM3886 without electrolytic feedback capacitor and (hardly) no DC offset voltage.
Marc.
 
Tom,
Of course I do know that!! But you have to read my earlier thread. The purpose of the extra resistor of 27 kohm is to cancel out the input offset voltage. That's why the value of this resistor must be determined experimentally. In one amplifier the value is 27 kohm and in the other 18 kohm. By these values the output offset voltage is less than 1 mV. If you make the value of this resistor zero ohm (which is the normal situation), then you are right.
So I was not lucky: this is the outcome of a well-thought-out idea.
But please read my earlier thread. And try to imagine the effect of the extra resistor of 27 kohm.
Marc.
 
Please link to your thread.

Cancellation of offset by adding impedance in series with the input is a common trick. Unfortunately, it requires good matching between the input bias currents to work well. The matching isn't bad on the LM3886 (10 nA, typ.; 200 nA w/c), but still that means your 18-27 kΩ resistor may add as much as 5 mV of additional offset.

You can get it to work once and if you're willing to tweak every implementation, you can get it to work more than once, but it's not a production solution.

These kinds of trick have been around since opamps were invented. You can find a good comprehensive list and description in Chapter 5 of Franco. I'm sure Jung covers it as well.

Tom
 
Hello TioFrancotirador and Tomchr,
There is a very very easy way to get rid of the feedback cap and have an offset voltage of almost zero Volt. In the two LM3886 amplifiers I build, one amplifier has an offset voltage of 0,2 mV and the other of 0,9 mV. These offset voltages are very stable under unstable circumstances like varying output power and changing temperatures.
See: LM3886 without electrolytic feedback capacitor and (hardly) no DC offset voltage.
Marc.

Thanks Marc,
I saw this thread before, however did not try it, since I achieved only a dozen of mV without cap and without the mod you described.
Anyways after building Modulus 86 I see no point building any "bare" based LM3886 designs :). Modulus 86 is relatively not that much expensive than LM3886DR and it is much, much, much better sounding with the DC offset of around 0,2mV out of the box.
 
Tom,
Indeed the input bias current through the resistor gives an offset voltage but that offset voltage is opposite to the normal input offset voltage. They both cancel each other, giving a zero output offset voltage which is - in the case of the LM3886 - very stable.

Indeed you have to tweak every implementation but that is also normal in many factory amplifiers where the output offset voltage and/or the quiescent current through the end transistors have to be adjusted.

This solution is not a trick. And this is a diyaudio (do it yourself) platform so what is wrong by tweaking?

Marc.
 
There is nothing wrong with tweaking. For some that is an exciting hobby. However, I deliver a production solution (build it and use it). I provide circuits which work well if built according to the instructions. All my circuits can be built using off-the-shelf parts and require no tweaking.

I understand how the circuit you propose works. I also understand its limitations. In my past two posts I attempted to make you aware of those limitations. I suggest you go read chapter 5 in Sergio Franco's book, "Design With Operational Amplifiers and Analog Integrated Circuits". The reason I keep recommending this book is that it is by far the best book on opamps I have come across. Secondly, I used to teach its content to senior (final year) undergraduate electrical engineering students at University of Washington, so I'm intimately familiar with it. Jung's book is harder to read, but is a free download from ADI's website.

Inserting 18-27 kΩ in series with the amp input like you propose might work in your case, but it comes at the expense of increased noise and lower bandwidth. I would also check the performance near clipping as you reduce the effectiveness of Cc (cap between the inputs of the LM3886). For these reasons I cannot recommend the tweak you mention.

Tom
 
Last edited:
I built a 4 channel LM3886 power amp which I use to power some bi-amp's speakers I built, and I am very happy with it, and am now in the process of building another 4 channel 3886 amp. The one thing I necessarily would have done different on you amp, is I'd lose all the connectors inside the chassis. Connectors can become weak and dysfunctional over time. Why not just solder the wires at the amp PCB's and elsewhere?
 

Attachments

  • IMG_1776 wb.jpg
    IMG_1776 wb.jpg
    803.5 KB · Views: 1,016
  • IMG_1767.JPG
    IMG_1767.JPG
    298 KB · Views: 1,012
  • 3886 pwr amp and tonemeister preamp wb.jpg
    3886 pwr amp and tonemeister preamp wb.jpg
    387.4 KB · Views: 985
The one thing I necessarily would have done different on you amp, is I'd lose all the connectors inside the chassis. Connectors can become weak and dysfunctional over time. Why not just solder the wires at the amp PCB's and elsewhere?

There's no need to use the connectors, I didn't. The cost of having the available option is only a few cm^2 of board space. I just deleted them from my BOM when I placed my order. Others can use them if they want them.
 

Attachments

  • Mod86_Plate.jpg
    Mod86_Plate.jpg
    622.2 KB · Views: 982
Marc,

The schematic shown in your thread is only good for canceling one polarity of DC offset. If the offset is of the other polarity, you'd need to move the resistor to the other (non-inverting) input. Your approach uses the op-amp's (and LM3886 is just a big op-amp) input bias current flowing through the resistor to create the cancelling offset. One can't assume that the change in DC offset with time and temperature will move in the same direction and change by the same magnitude as the input bias current. If input offset voltage and bias currents don't change in the same direction and by the same proportion with change in time/temperature, the amount of correction won't be correct. (Using resistors in the > 5k Ohm range affects noise as Tom pointed out)

If one wanted a non-servo offset correction circuit, the more traditional approach is to have a voltage divider (possibly with potentiometer) between Vs+ and Vs- and then connecting the voltage divider tap to the summing node (inverting input) with a resistor. This will give better results and avoid the noise problem, though still doesn't correct for changes with time/temperature.

The incremental cost to go with Tom's DC servo is minimal and it gives consistent performance over all changing conditions. Yes, as DIY, we can tweak and add cost that might not be practical in a commercial product. For exactly that reason, I'd put a couple extra bucks into a servo and gain a higher-performance solution
 
Marc,

If one wanted a non-servo offset correction circuit, the more traditional approach is to have a voltage divider (possibly with potentiometer) between Vs+ and Vs- and then connecting the voltage divider tap to the summing node (inverting input) with a resistor. This will give better results and avoid the noise problem, though still doesn't correct for changes with time/temperature.

Wouldn't this ruin the PSRR?
 
The one thing I necessarily would have done different on you amp, is I'd lose all the connectors inside the chassis. Connectors can become weak and dysfunctional over time. Why not just solder the wires at the amp PCB's and elsewhere?

You can certainly solder the wires directly into the boards, but that does tend to make assembly harder than it needs to be.
Amp modules by companies such as Hypex, Pascal, et al. all use connectors. The Mega-Fit Molex connectors I use are found in nearly every PC and have been for years (if not over a decade). They're very reliable and are rated for 23 A of continuous current.

Wouldn't this ruin the PSRR?

Not if you bypass the reference voltage developed from the supply rails. If you want better PSRR, you can use a precision reference and divide that down. That's just another trick (or tool if you don't like the word trick) in your bag of tricks (or toolbox).

Tom
 
If one wanted a non-servo offset correction circuit, the more traditional approach is to have a voltage divider (possibly with potentiometer) between Vs+ and Vs- and then connecting the voltage divider tap to the summing node (inverting input) with a resistor. This will give better results and avoid the noise problem, though still doesn't correct for changes with time/temperature.

True. If a servo or feedback cap is to be avoided, what you describe is my favourite offset nulling method. The only drawback is that it requires adjustment/calibration during assembly. Not an issue in a DIY environment where time is free, but very costly in production (as I know you know).

The incremental cost to go with Tom's DC servo is minimal and it gives consistent performance over all changing conditions. Yes, as DIY, we can tweak and add cost that might not be practical in a commercial product. For exactly that reason, I'd put a couple extra bucks into a servo and gain a higher-performance solution

That's exactly the reason there's a DC servo in the Modulus-86 and -286. The LM3886DR is designed for the more cost conscious builders out there, thus, replaces the DC servo with an audio grade bipolar electrolytic cap. I'm not able to measure any contribution on THD, IMD, or any other parameter from that cap, so I'm really quite content with that solution.

Tom