Lower chipamp distortion through PCB routing

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
A focusrite 18i8 and a dedicated load. It's very similar to your setup actually, the load is kind of a simplified version of the Millet interface.

I use a scope to check the amp output levels. Most measurements are made so that the input going into the 18i8 is at -5dbfs (knowing that the 18i8 inputs max out at +16dbu).
 

Attachments

  • IMG_20180608_163106930.jpg
    IMG_20180608_163106930.jpg
    114.8 KB · Views: 719
Member
Joined 2009
Paid Member
A focusrite 18i8 and a dedicated load. It's very similar to your setup actually, the load is kind of a simplified version of the Millet interface.

I use a scope to check the amp output levels. Most measurements are made so that the input going into the 18i8 is at -5dbfs (knowing that the 18i8 inputs max out at +16dbu).

Thank you.

The setup is indeed quite similar, as are the specs of the soundcards, yet your noise floor in the spectrum atatched to your post above is 10 to 20dB better than I can get from my soundcard connected to itself, without the interface. Curious.
 
Member
Joined 2009
Paid Member
Are you certain your soundcard is operating in 24 bits ? Windows can be tricky...

Thank you - for the second time, you point me in the right direction :)

The card used 24 bits, but ARTA used 16-bit dithering. With 20 bit dithering, the picture is somewhat better - see attached. This is just the soundcard by itself.
 

Attachments

  • img.png
    img.png
    60 KB · Views: 665
Last edited:
Member
Joined 2009
Paid Member
And here are the spectra of the amplifier as before: simple unregulated power supply, 18.5Vrms (about 43W) into a 8ohm resistor.

Here I am using Pete Millett's soundcard interface, but switched from ARTA to SpectraPLUS for FFT processing.
 

Attachments

  • 2018-06-30 (4).png
    2018-06-30 (4).png
    331.8 KB · Views: 653
  • 2018-06-30 (5).png
    2018-06-30 (5).png
    349.3 KB · Views: 638
Member
Joined 2009
Paid Member
Inverting vs. non-inverting vs. differential configuration

I got some questions regarding the best way to do the grounding of an amp based on my board, and based on that, it makes sense to explain the differental chipamp configuration used by the board.

The datasheet recommends, and virtually everyone uses, the non-inverting configuration - see the top left corner of the attached illustration. I rearranged it to make it clear to myself how it works - see the middle left block - so maybe other find it helpful, too. In this configuration, the opamp (such as an LM3886) strives to minimize the voltage between points A and B by driving its output appropriately. A is the input signal, referred to GND. B is the output signal divided by R6/R7, also referred to GND. it immediately follows that, for a proper amplification, A and B must refer to the same GND - that is, the three GND points (the input reference, the power ground/output reference, and the bottom end of the R6/R7 divider) must be one and the same point. This may become problematic if these points are in fact physically separate, such as in a multichannel system. Separating the grounds will generate all kinds of problem, commonly called ground loop issues, for which there are more or less effective remedies and workarounds. A subtler problem is that the input capacitance of the opamp is non-linear and changes with input signal, introducing distortion via nonlinear input current - some call it common mode error.

The common mode error is dealt with in the inverting configuration (bottom left block on the illustration), where the voltage between A and B is always zero and the input capacitance plays no role. Because of this, the inverting configuration is almost always used in high speed circuits. However, it features low input impedance (equal to R10) and suffers from the same single-point-of-reference problem as the non-inverting. With the LM3886, there is one more issue: if the output is left floating, the gain of the circuits becomes unity, and LM3886 is unstable when the gain is below 10. Because of that, every circuit using LM3886 in the inverting configuration drives it with some buffer with low output impedance - see this one, for example.

My board uses a differential configuration (right column on the illustration). Here, the input reference and the output reference are completely separate and do not need to refer to the same point at all. The input may be left floating or grounded at either end, it makes no difference. You obviously need to ground the output because that is where the power ground is connected. Input capacitance plays no role because the charge currents for the input capacitance cancel. Quoting Bruno Putzeys in his article referenced in my first post, "Whoa. Not only does differential circuit design do away with current loop problems, it actually eliminates a significant source of distortion. If panaceas exist, this must be one of them". Differential input impedance is high, as is the input impedance with the inverting input grounded, although not as high as one can have in a non-inverting circuit. You still need to either ground or drive from a low impedance source the inverting input for LM3886 stability.
 

Attachments

  • inverting non-inverting.png
    inverting non-inverting.png
    43.9 KB · Views: 622
  • Thank You
Reactions: 1 user
Very interesting, great results!

Did you try your boards with regulated PSU?
Things would improve further with regulated PSU on bare LM3886.
I was thinking of implementing some kind of basic filtering on PSU section of my LM3886 board I play to play with.

CRC and CLC come to mind with CLC being very effective with HF filtering but both options increase power supply output impedance by R component that possibly might affect LM3886 in a negative way since it is a class AB(I actually have not done any measurements, but I would be very interesting how output impedance of PSU affects performance of the chip).

Even CLC is not without problems since some R component is needed in L to stop the low pass filter peaking around corner frequency.
 
Member
Joined 2009
Paid Member
Did you try your boards with regulated PSU?
I have not, and it's not on my to-do list for now. Although the amp may benefit from a low output impedance power supply, I believe building a high current regulator with consistently low output impedance across the audio range is non-trivial. In any regulator, an error amplifier tries to force the output up or down by comparison with an internal reference. As the error amplifiers gain falls off with increasing frequency, so too does the regulators accuracy and output impedance. See, for example, the output impedance of an LM117/317, taken from its datasheet. Moreover, the impedance that rises with frequency makes the regulator behave like a synthesised inductor, which forms resonant circuit with downstream capacitors, with effects on audio performance.

Filtering sounds like a no regret move, and I used CLC filters (heavy and bulky) in low/no-global-negative-feedback amplifiers, with clearly audible improvement. Douglas Self in his book recommends in particular filtering the rail which is used as a reference in the second (transimpedance) stage of power amplifiers.

By joining the -IN to the star point instead of the OUT G you added some milliohms resistance made of the strip linking the OUTG to the ground.
There is no star ground in this amplifier - it is not needed with this topology, unlike the standard non-inverting one. Feedback is taken directly from the OUT/OUTG wire pads on the board.

That said, varying output impedance of an amplifier can be an interesting thing to try - see here.
"The effects are not subtle... There was no consensus about what sounded best, either - some preferred an all-out current source, others a voltage source."
 

Attachments

  • LM117 Zout.png
    LM117 Zout.png
    92 KB · Views: 317
Member
Joined 2009
Paid Member
my real circuit diagram

As I tried to explain earlier in this thread, I used a differential topology, not the usual non-inverting one - see the attachment. There is no star ground, no implicit current feedback and no reliance on the impedance of wires or traces.
 

Attachments

  • what it is and isn't.jpg
    what it is and isn't.jpg
    79.7 KB · Views: 515
Member
Joined 2009
Paid Member
Pin 7 looks like a star ground on the PCB but is not a star ground in the sense it is not used as a reference point for any signal. (Pin 7 is the return pin for the muting current).

The commonly used non-inverting topology relies for its proper functioning on several points in the circuit (bottom ends of the source, the load, and the feedback divider in the schematic in my post above) having exactly the same potential. Rtrace, although small, is always greater than zero - on my PCBs, the calculation gives about 2mOhm. That is 0.025% of the nominal load impedance, and its effects can be measured and heard.

If I understood correctly, kokoriantz makes the point that trace resistance can be used to provide some negative feedback based on the output current, which would increase the output impedance of the amplifier but make it less frequency dependent, which in turn would improve the sonics.

I find this idea very interesting, but my board is not relying on any trace impedance to be zero, and thus such negative feedback doesn’t exist on the board - which is one reason for the its low distortion.
 
Last edited:
Be at least honest.

Well. You asked. :)

It looks like you're missing at least one decoupling cap per rail. Also, the decoupling for the VEE pin is quite far away, thus not all that effective. I suggest taking a look here: Taming the LM3886 - Supply Decoupling.

Your supply routing is about the worst you can do for rail-induced distortion. You want the supply lines to be tightly coupled and low impedance. That means wide pours that come in side-by-side from the left (as the supply pins are on the left side of the chip). Douglas Self writes about that in his Power Amp Design book (and Wireless World article series that became the book).

You'll want to run your loop analysis all the way past the unity-gain frequency. Also, I suggest using a reactive load. At least run a few test cases with 8 Ω in parallel with 1 nF, 10 nF, 100 nF ... or better yet a simulated speaker load, such as the model Stereophile is using.

Looks like pin7 is the star ground.

The star ground layout is great for controlling current flows. Unfortunately, it also features the largest ground impedance of any layout, which means you get the largest error voltages developed across the ground with that type of layout. To minimize the error voltages, you need to minimize the ground impedance (i.e. use a ground plane). The ground plane makes it harder (but not impossible) to control the current flows, so you will have to be more careful with your component placement to ensure that sensitive nodes are located in a quiet spot of the ground plane. An alternative is to use multiple ground planes and join them at the ground reference point. You can read more about grounding here: Taming the LM3886 - Grounding.

I think the point that is being made is that input and output grounds should be connected so that Rtrace is zero.

Yep. I'd expand that to Ztrace should be zero. I.e. minimize the inductance as well as the resistance. That's why I like planes and pours.

Tom
 
(Pin 7 is the return pin for the muting current).

It's actually the reference voltage used during mute. The output of the chip is driven to V(pin7) during mute. The voltage on pin 7 is not used for anything else according to the equivalent circuit schematic of the LM3886.

If I understood correctly, kokoriantz makes the point that trace resistance can be used to provide some negative feedback based on the output current, which would increase the output impedance of the amplifier but make it less frequency dependent,

I would be really careful about using trace resistances for anything that involves stability or performance. The resistance of the board traces is generally not that tightly controlled.

which in turn would improve the sonics.

[citation needed] ;)

I find this idea very interesting, but my board is not relying on any trace impedance to be zero

Good! Because you'll never get zero. Unless you can find a board vendor who can make boards with a layer stack that's superconductive at room temperature. :)

Tom
 
Star ground as being the point where power, input and output grounds connect together (pin7).
Where it is different than most is that the feedback resistor is connected to the input ground and the input resistor connects to the output ground.

I have been using a similar approach for a while now but the layout has a single point at the output.
 
Star ground as being the point where power, input and output grounds connect together (pin7).

I understand that. It's still not a good idea.

Where it is different than most is that the feedback resistor is connected to the input ground and the input resistor connects to the output ground.

OK. The feedback current is not pretty. It contains all the error signal from the chipamp itself. From your description, it sounds like this error is then added to the input signal. That's not a good idea.

I have been using a similar approach for a while now but the layout has a single point at the output.

Which is usually where you want it.

Tom
 
Its very easy to get caught out with pcb design.
I got caught out recently with a ICL7660 voltage inverter for -5 volts.
I hadn't realised it oscillated externally and the 10KHz was getting into my audio signal.
I had run an audio signal underneath the 7660's and it caused havoc. I had to cut the track at both ends and link it direct.

The biggest problem I find is mixing power supply ground in with audio ground.
Its vital they are kept separate and only connect once at pcb edge connector.

Routing audio near high AC volts is out too as it easily picks up hum.
Also routing audio next to power transformers isn't good.
 
Member
Joined 2009
Paid Member
Re supply decoupling

Tom,

Have you, by any chance, analyzed how supply decoupling specifically affects the performance of LM3886? Your page on the topic starts with the assertion that if the supply to the LM3886 is bouncing, the LM3886 may behave in unintended ways, but you never give any design parameters. You do mention, however, that no power supply non-idealities are included in the LM3886 model, and that the data sheet is ambiguous on the exact requirements for bypassing. As a result of this ambiguity, many people here on the forum like to pile up multiple caps on each rail, although the version with one cap per rail works remarkably well. In your view, how much decoupling is enough, and what specifically are the ill effects of various kinds of insufficient decoupling?
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.