Reference ABX hardware design and implementation

[mcdn]

Edit 2026-07-31, see progress update at https://www.audiosciencereview.com/...-design-and-implementation.72589/post-2649465

-----------------------

This is a new thread to take detailed design discussion out of https://www.audiosciencereview.com/...ifferences-are-finally-proven-with-abx.72444/

I have a design sketched out and modelled but would like some feedback on features before locking it in, routing the board and ordering a prototype. It's single ended for the obvious cost reasons, and split between the main line level board and a separate power switching board. Single OPA1612 at this point as the best balance between noise, THD and complexity. Analog switches instead of relays - TI TMUX7612. PCM1808 ADC for calibration. An MCU for calibration and control - I like the Picos, but ESP32 is also an option.

Audio performance

• Maximum line input/output level: 4 Vrms at unity gain
• Input impedance: 10k
• DUT gain matching: 0 to +12 dB per channel
• Gain-matching precision: Better than 0.02 dB
• Channel-matching precision: Better than 0.02 dB
• End-to-end THD+N: Better than −130 dB(A)
• Output noise: Less than 0.63 µVrms(A)

Features

• Automated whole-path calibration — Measures the complete listening path and matches DUT levels using any audio signal.
• Temperature-controlled audio path — Stabilises critical analogue circuitry for consistent performance.
• Cueless switching — Silent solid-state line switching and a controlled output-mute interval conceal which device is selected.
• Line-level A/B/X — Compare two stereo sources, DACs, preamps or other unbalanced line-level devices through one common stereo output.
• Power-amplifier A/B/X — Compare two amplifiers through the same loudspeaker, including conventional, bridged, floating and compatible class-D outputs.
• Loudspeaker A/B/X — Compare two loudspeakers from one amplifier, with independently stored level settings established using an SPL meter.
• External power-switching interface — Use the matching power-relay board or connect a compatible user-provided alternative for coordinated external equipment switching.
• Selectable AC or DC coupling — Headers on board can be used to bypass AC coupling capacitors

Operating requirements

• Line load: 2 kΩ minimum; 10 kΩ or greater recommended
• Ambient temperature: 18-25 °C recommended for optimal performance; operation supported up to 35 °C
• Amplifier/speaker switching load: Up to approximately 500 W, 80 V and 20 A
• External power: 48 VDC, 1 A regulated Class II adapter

 

[voodooless]

I’m following this!

Will the amp and loudspeaker switches also be TMUX7612? They don’t seem to be particularly suited for this with only 470mA current and 1.1R resistance.

 

[mcdn]

Basic principles, as suggested by @KSTR:

• @mcdn is the product owner - the product owner has final say over all decisions but can't be the expert for everything or even anything
• Community contributions are needed for technical, logistical and emotional support!
• We will move fairly rapidly from scoping through design to prototyping, without compromising on quality or missing essential community and user feedback

 

[mcdn]

I’m following this!

Will the amp and loudspeaker switches also be TMUX7612? They don’t seem to be particularly suited for this with only 470mA current and 1.1R resistance.

No, provisionally Panasonic AHES4291: https://jlcpcb.com/partdetail/PANASONIC-AHES4291/C719196

 

[voodooless]

No, provisionally Panasonic AHES4291: https://jlcpcb.com/partdetail/PANASONIC-AHES4291/C719196

You may consider something like AQZ202 as a solid state, silent alternative.

 

[mcdn]

Priorities…

Now:

• Input impedance, AC coupling decisions. provisional plan is 10k input impedance with a 47u coupling cap by passable with a jumper. Coipling cap choice is important, provisionally Nichicon UES1C470MPM (https://jlcpcb.com/partdetail/Nichicon-UES1C470MPM/C1669286)
• Other spec decisions, led by community queries
• Consideration of adding consistent sounds during switching instead of silence as proposed by @Josh83 in private message
• Investigation of solid state relays for power switching to eliminate clicking

Later

• Form factor
• Testing and validation
• Control interfaces

 

[mcdn]

You may consider something like AQZ202 as a solid state, silent alternative.

My initial observation is that it's only rated to 3A and 48V, and has non-negligible on-resistance, but another solid state relay could be good. My design goal is 60V and 20A (not necessarily at the same time! we need to handle both high and low impedance speakers)

 

[voodooless]

Input impedance, AC coupling decisions. provisional plan is 10k input impedance with a 47u coupling cap by passable with a jumper

Make it switchable as well You could even add some bits to ABX coupling caps

 

[mcdn]

Make it switchable as well You could even add some bits to ABX coupling caps

Very amusing

 

[mcdn]

Make it switchable as well You could even add some bits to ABX coupling caps

Seriously though, I'm not yet convinced AC coupling is required. It does mitigate thumps due to DC offset differences between DUTs, but with output muting between selections those won't be audible anyway.

 

[KSTR]

What would be the planned analog control elements for the gain/level trims? I don't see it mentioned yet, and 0.2% steps (with +-0.02% or so tolerance) is a tough call.

 

[KSTR]

Seriously though, I'm not yet convinced AC coupling is required. It does mitigate thumps due to DC offset differences between DUTs, but with output muting between selections those won't be audible anyway.

Assume a zero DAC input signal, but presented with different analog DC offsets. How would you mitigate against the additional and different clicks (unless both offsets are exactly zero) on mute/unmute?

 

[pma]

Capacitors would not help. There would always be a charge mismatch (for various reasons) before/after switching and an audible click. I am not sure that nit-picking in details would result in a universally usable instrument. Nit-picking in details may mean overlooking the complete test requirements.

 

[welwynnick]

What would be the planned analog control elements for the gain/level trims? I don't see it mentioned yet, and 0.2% steps (with +-0.02% or so tolerance) is a tough call.

Can I suggest using an identical pair of external high quality pre-amps ?
Yes, they're not as good as a resistor network in theory, but they do have a few advantages:

• Available off-the-shelf
• High input impedance
• Low output impedance
• Known performance
• Better bandwidth, noise and distortion than any source.

 

[pma]

Can I suggest using an identical pair of external high quality pre-amps ?

From practical point of view, I think you are right.

 

[mcdn]

Can I suggest using an identical pair of external high quality pre-amps ?
Yes, they're not as good as a resistor network in theory, but they do have a few advantages:

• Available off-the-shelf
• High input impedance
• Low output impedance
• Known performance
• Better bandwidth, noise and distortion than any source.

The issue is repeatability. A dedicated box gives assurance that every test meets a common baseline, and while it can't quite eliminate reviewer error it can get close. It can also provide an audit trail.

 

[mcdn]

Here are more technical details that will probably answer some questions but then also generate more, which is great!

The design keeps line input/output and control separate from power signal switching. This allows for both signal and supply separation, as well as catering for users who only want the line capability. Honestly the power switching part is pretty simple compared with the line level stuff.

Fundamentals

The line/control board contains the stereo inputs and outputs, switched gain, measurement ADC, controller, local power supplies and thermal control. Each channel is conceptually:

Input A ─┐
         ├─ input selector ─ OPA1612 gain stage ─ output switch ─ line output
Input B ─┘                         │
                                  └─ switched feedback ladder

The design is ground-centred and runs from bipolar ±15 V rails. The normal listening path is direct-coupled, with AC coupling available by removing a jumper. A 47 Ω output-isolation resistor sits outside the feedback loop, and the normal and measurement RCA outputs are wired in parallel after the output switch.

The OPA1612 is non-inverting and has a fixed 180 Ω feedback resistor and twelve branches from the inverting input towards ground, giving fine gain adjustment from 0 to +12 dB. The branches are in four nested three-bit banks. The first is always visible; normally open switches progressively connect the others as needed. This keeps the lowest-resistance, highest-capacitance branches out of the way in the fine-step, low-gain region. If you're comparing two DACs with nominal 2Vrms output you may only need the first bank. That said, this is still an area of significant design risk.

The gain network around each op-amp looks like this:

                         OPA1612
Selected input ───────────|+\
                          |  \_____o──► output
Feedback node o───────────|− /     │
              │           | /      │
              └────── 180 Ω ───────┘
              │
              ├── bits 0–2: 3 × (R + SW) ── GND
              └── range SW 1
                   ├── bits 3–5: 3 × (R + SW) ── GND
                   └── range SW 2
                        ├── bits 6–8: 3 × (R + SW) ── GND
                        └── range SW 3
                             └── bits 9–11: 3 × (R + SW) ── GND

During calibration, one PCM1808 input measures the complete left and right signal chains at the final line output. The MCU stores the best resistor codes for each gain level. Calibration therefore covers resistor error, amplifier gain, analogue-switch resistance, connectors and routing loss, without requiring precision resistors.

Input switching

When switching inputs the idea is to use a fixed sequence of operations to try to eliminate cues. Ticks refer to 1ms MCU control loop runs, so 10ms total, which is maybe not standard but obviously can be adjusted in software. Total cue elimination is an open question for community feedback.

Tick 0     Open both line outputs
Tick 1     Open both source paths and calibration taps
Tick 2     Apply the complete stereo gain state
Tick 3     Connect exactly one stereo source
Ticks 4–9 Make no analogue-path changes; allow quiet settling
Tick 10    Reconnect both line outputs

All control is provided by an MCU with GPIO expanders as needed. Displays, buttons and other user interface controls are not scoped yet.

Thermals

The analogue switches need to be kept near 25 °C for optimal on-resistance linearity, so the MCU also controls a pair of heater resistors. This is a simple on/off loop using a thermometer IC, and is not switched during ABX testing.

Grounding

Audio, ADC, digital and power returns are treated as distinct current-return regions joined only at controlled points. Converter switching loops and heater current remain outside the audio region, and control traces do not enlarge the sensitive feedback-current loops.

Power-amplifier and loudspeaker switching

High-current switching is kept on a separate board. It switches both conductors of each amplifier path. Amplifier and speaker current never travel through the line/control-board harness.

The provisional design uses mechanical relays to support both high-voltage/high-impedance amplifier and speaker combinations and vice versa, giving around 80 V and 20 A as the design target.

Mechanical relays do introduce some risk of audible cues. Proposals to mitigate this are welcome!

 

[ATTI]

From the point of people who hear (or beleive in) the difference for example in op amp sounding, such setup is incorrect in its basis, this device should be without any parts in signal path except the same connectors and the same relays.

 

[ppierre]

Mechanical relays do introduce some risk of audible cues. Proposals to mitigate this are welcome!

Simply use two relays, on used to switch, one dummy to simulate switching. X switching would always switch a relay, the used one or the dummy one.

Make this a separate box with a computer input. And it should suffice for some ABX testing.

 

[mcdn]

From the point of people who hear (or beleive in) the difference for example in op amp sounding, such setup is incorrect in its basis, this device should be without any parts in signal path except the same connectors and the same relays.

The opamp itself is almost the least distorting or noise generating part of this circuit. Resistors and switches (and their associated capacitance), add more noise and distortion than the OPA1612. Opamps do measure differently, that's why the specific choice of opamp here is so important. You also have JFET vs Bipolar vs CMOS inputs, each with different tradeoffs. You absolutely could not put an NE5532 in here and expect the system to operate correctly.

So yes, some people will always believe op-amps have a "sound". The sound of this one in this circuit is buried at -130dB or more below the signal.