Reference ABX hardware design and implementation

[KSTR]

Btw, looking at your preliminary layout, did I just spot a typical error I see that can easily spoil signal integrity? A decoupling cap left and right of the OPA1612, with each one having its own local GND via to the plane. A big no-no. Combine the cap GNDs first, then via down into the plane.

Otherwise you'll have different GND paths for each half-wave of load current (once it's above class-A bias region).
And why does the OPA1612 have a thermal pad when it is SO-8? Only the WSON package has that. That pad gets in the way of the above, as the GND via(s) for the decoupling return should ideally be right under the chip.

EDIT: Or are the two smaller chips at the bottom end two 1612's in WSON? If so, same problem as far as I can see.

 

[mcdn]

Btw, looking at your preliminary layout, did I just spot a typical error I see that can easily spoil signal integrity? A decoupling cap left and right of the OPA1612, with each one having its own local GND via to the plane. A big no-no. Combine the cap GNDs first, then via down into the plane.
View attachment 553127
Otherwise you'll have different GND paths for each half-wave of load current (once it's above class-A bias region).
And why does the OPA1612 have a thermal pad when it is SO-8? Only the WSON package has that. That pad gets in the way of the above, as the GND via(s) for the decoupling return should ideally be right under the chip.

EDIT: Or are the two smaller chips at the bottom end two 1612's in WSON? If so, same problem as far as I can see.

Superb catch sir, were you playing in goal for Cabo Verde at the World Cup?

I'm not sure what's up with that layout, but here's the latest one. We switched to the OPA2211 a few pages back-thread, which has the more complex package. I haven't hand-optimised it yet apart from applying the joined bypass grounds, so don't pay attention to silly kinks or over-close courtyards, but it's electrically passing.

• F.cu (red) is analog signals
• In1.cu (green) is AUDIO_GND,
• In2.cu (orange) is V13_N
• In3.cu (cyan) is V13_P
• in4.cu (pink) is DIGITAL_GND
• B.cu (blue) is mixed.

[more recent image uploaded, apologies if you were looking at the older one]

 

[somebodyelse]

The later shows that XLR input is in fact not balanced, it uses 2 identical paths to drive the bridged amplifier circuit. When it is driven to one half as in case of RCA-XLR adapter, it gives only 1/4 maximum output power. Such design is heavily compromised and non-standard.

I'm suspicious that some of the TPA325x amps with 'balanced' inputs may be doing the same thing. I vaguely remember someone reporting results similar to your B100 testing with one, but I don't remember which one, and I don't think it's been checked on many others.

 

[mcdn]

I'm suspicious that some of the TPA325x amps with 'balanced' inputs may be doing the same thing. I vaguely remember someone reporting results similar to your B100 testing with one, but I don't remember which one, and I don't think it's been checked on many others.

I think most TPA3255 modules just follow the TI reference design. I’ve measured at least 4 different ones and they all did balanced inputs properly. No significant difference to SE at the amp output of course, but the correct pair of opamps has been there on every board.

 

[mcdn]

Hey folks! Sorry for the long gap between updates. I've been collaborating directly with @KSTR because he's designing a fully-balanced version of the line-level module. Having a genuine option to drop in has shown a lot of places where the existing design wasn't modular enough. There are still some things to settle out but concrete changes so far include:

• Sharing the protected DC input directly with modules. The DC input will now be strictly 12V, not a range from 12-24V as before, but downstream modules can draw up to 10W from V12_PROTECTED
• Moved from +-15V isolated to +24V isolated as the primary supply, with audio rail generation on each module that needs it. Traco THM 10-1215 is the selected part, delivering 10W 24V isolated with 20pF isolation capacitance and 1% regulation. This puts the isolation in one place, simplifying downstream power designs.
• A consequence of the change to the power architecture is that we can remove some digital isolation barriers. Isolation remains in place on the USB audio input and coaxial SPDIF, but is no longer needed on the I2C and I2S lines.
• A switch of component for the fading LDR (light dependent resistor) to allow 0.5dB steps all the way down the fading ramp to -60dB without some hacks that were in there before. The new part is much more linear at the low end of the range. (https://jlcpcb.com/partdetail/Senba_SensingTech-LCR0202/C82366)
• We found a better choice of small signal relay - Omron G6K series. Same footprint but enough of a higher coil resistance that we could eliminate some driver parts, using the existing LP5009 LED driver directly (with a small RC filter)

Physically we've established that the digital module will fit in the chosen case with either the SE module or the balanced module, but the back panel layouts will of course be different.

The speaker module remains in a separate matching enclosure, linked via a DE-15 connector for power, control and monitoring. The "silent switching" mosfet based speaker board has seen a switch to the TPSI3050 gate drivers, based on availability and simplicity. The MOSFETs remain the IPT025, with a 150V Vds. The mosfet board is therefore specced at:

• 160V P-P - enough for a Hypex NCx500 at full power
• 10A rms
• 20A peak

 

[pma]

We found a better choice of small signal relay - Omron G6K series.

I hope you will have better experience with these parts than I have had, after about 2 years of operation I had to remove them all. My experience has been much better with G5V-2 series.

 

[mcdn]

I hope you will have better experience with these parts than I have had, after about 2 years of operation I had to remove them all. My experience has been much better with G5V-2 series.

The coils will run at 1/4 power after activation, so that should help. The G5V-2 are definitely more robust but are almost 4 times the board area and 5x the power requirement.

 

[paschulke2]

I hope you will have better experience with these parts than I have had, after about 2 years of operation I had to remove them all. My experience has been much better with G5V-2 series.

What was wrong with the G6K relays after 2 years? Mechanical Problems? Bad contact at low voltages or currents (signal level ...)?

 

[pma]

The later. My favorite is Axicom D2n.

 

[Sokel]

Why not reed relays?
10 times faster, at least and essentially unlimited lifespan.

(and one less tell, as they are essentially silent inside a case, one has to put its ear cm from them to hear the faint *ting*.

 

[bigLP]

Why not reed relays?
10 times faster, at least and essentially unlimited lifespan.

(and one less tell, as they are essentially silent inside a case, one has to put its ear cm from them to hear the faint *ting*.

I have a slightly adjacent question I would like to ask quickly, not to derail thread. Back when I was a Naim guy they used reed relays. They failed in one channel of two different Nait XS I owned. I seem to recall them saying they got a bad batch or something. Are reed relays usually reliable and why use them? I am technically limited, I googled it but didn't learn anything,

 

[marin.weigel]

... I haven't hand-optimised it yet apart from applying the joined bypass grounds, so don't pay attention to silly kinks or over-close courtyards, but it's electrically passing.

Sorry for my injecting, just lurking here and not following every post, but I see issues with the layout for the decoupling caps - please ignore at your discretion!

For them to be maximally effective, one should avoid connection to the respective power pin through vias, due to their intrinsic inductance, but rather connect with an as short as possible trace directly to the respective pin. I had been taught this being more important than the return path side.

 

[Sokel]

I have a slightly adjacent question I would like to ask quickly, not to derail thread. Back when I was a Naim guy they used reed relays. They failed in one channel of two different Nait XS I owned. I seem to recall them saying they got a bad batch or something. Are reed relays usually reliable and why use them? I am technically limited, I googled it but didn't learn anything,

C.P. Clare used to be a gold standard about quality, I have some of them working for 20 years at a DIY el. x-over and a bunch of them still new for future projects.
Reed relays usually fail when used outside their envelope, I wouldn't use tham at a power amp for example.

But for small signal, pres, x-over, etc, it would be my first choice.

 

[mcdn]

Sorry for my injecting, just lurking here and not following every post, but I see issues with the layout for the decoupling caps - please ignore at your discretion!

For them to be maximally effective, one should avoid connection to the respective power pin through vias, due to their intrinsic inductance, but rather connect with an as short as possible trace directly to the respective pin. I had been taught this being more important than the return path side.

Nanohenries at most, totally irrelevant at sub MHz frequencies

 

[PristineSound]

I don't recall the relays used in the Benchmark preamp, but supposedly, they lasts forever, according to the manufacturer.

 

[mcdn]

Why not reed relays?
10 times faster, at least and essentially unlimited lifespan.

(and one less tell, as they are essentially silent inside a case, one has to put its ear cm from them to hear the faint *ting*.

Physically much larger, especially in DPDT configurations (or if single pole, twice the part count), expensive, through hole instead of surface mount. The quiet operation is a potential bonus, but miniature mechanical relays are pretty quiet too, and we don't need the extra speed in this application.

 

[KSTR]

I hope you will have better experience with these parts than I have had, after about 2 years of operation I had to remove them all. My experience has been much better with G5V-2 series.

Bulk failures like that suggest contact contamination from improper storage or soldering, or PCB cleaning. Notably for SMT types.
I've used G6K many times (but only the THT variants) and no bulk failures so far, only the very occasional lemon.

I agree AIXCOM/TE is also very good and often used for audio switching, and they have 100mW and even 50mW high sensitivity types in their IM06 range.

 

[KSTR]

Sorry for my injecting, just lurking here and not following every post, but I see issues with the layout for the decoupling caps - please ignore at your discretion!

For them to be maximally effective, one should avoid connection to the respective power pin through vias, due to their intrinsic inductance, but rather connect with an as short as possible trace directly to the respective pin. I had been taught this being more important than the return path side.

For the typical audio opamp with about 100MHZ GBWP, via inductance is irrelevant and other things are more important. Notably that return currents are combined before entering the plane, and loop area is minimized so that class-B currents don't pollute the GND and don't couple inductively into sensitive nets.

 

[marin.weigel]

don't couple inductively into sensitive nets

Restricting the ground plane around sensitive nets is a layout best practice as well in such circumstances.

 

[KSTR]

Well, I've found the best implementation (for opamps) in ultra-critical applications is to route the chip's supply tracks back to the center of the opamp following the lead-frame layout, then via-through to the supply planes. I'm often using "augmented plane capacitance" topology rather than strict local decoupling which allows to route proper GND and supply plane "fingers" to the chip to 100% confine the three current paths and minimize the emitting loop area, with the decoupling placed where it is convenient (still close to the chips).
But often a simpler way (as shown in my post #341) is good enough, the main point always being the combination of the return currents, and small loop area before combining.

Plane keepouts are usually required when one needs to minimize capacitve loads. Plane peninsulas help to avoid unwanted currents flowing across a section of PCB with a GND reference node.