I honestly don't get it... are we saying up to now every ABX test methodology has been flawed? So much so we need fundamentally new tech?
-
Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!
Reference ABX hardware design and implementation
- Thread starter mcdn
- Start date
This might be to be discussed ... just get some more popcorn ...
Depends on the sensing strategy chosen. Balanced sensing at the source (a DAC's unbalanced output) takes shield-current-induced voltage drop out of the signal path, and shield current could be reduces with hybrid grounding of the shield for a best compromise, but it would required a balanced ABX box input (not rocket science either, like shown in my post here).
Of bigger importance in general, IMHO, (also @mcdn), is a low mains leakage supply for the ABX box, and standard class-II SMPS brick is the worst-case choice (the dreaded EMC cap of a few nF bridging primary to secondary side, and the primary potential being full- or half-wave rectified mains voltage, ouch!).
I'll never use anything else again than isolated DC/DC converter modules with medical leakage rating, fed from a class-II supply (again ideally one with medical leakage rating) whose output GND is connected to the common "zero signal ground plane" in signal-integrity lingo.
The official requirement of a few 100ms of gap for the switchover is a known flaw disturbing focused perception. Hard muting is not required, that is fade-out --> silence --> fade-in is allowed, but direct cross-fade is not. Silence might be replaced with cross-fades to/from any other signal as long as it doesn't induce cues. Pink noise, envelope-following the signal level, works best IME, disrupting auditory flow as little as possible.
OP
- Joined
- Mar 7, 2020
- Messages
- 1,203
- Likes
- 1,917
- Thread Starter
- #86
There are a lot of "OK" solutions out there, but most of them have flaws that make them hard to trust as definitively removing all cues. Take gain matching for example - we know 0.2dB is an audible difference for sure. It's probable that trained listeners can detect a 0.1dB difference. So for starters any solution has to do better than that, and do it reliably. Listening to a software-based ABX test as with foobar through headphones is a known-good protocol, so that's the level of assurance we'd like.
How about off grid, when a nice big battery bank is feeding the entire system?
Nice quality SW inverter(s) a necessary evil, hope to only use for line-level gear, nothing thirsty.
Can you reco DCDC lines that go for reasonable prices?
I follow Vicor sales on eBay, but a bit above my pay grade, I prefer finished products in cases.
General tips appreciated too, whatever comes to mind.
For example, miniDSP C-DSP 8x12 is apparently quite RF/EMI sensitive. Upstream bank is likely nominal 24V but might get up to 48Vdc
How about something like this:
External medical grade "brick" for mains -> 24V DC. It can live outside the box.
And inside, a medical grade isolated 24V -> +/- 15VDC. Audio ground is made between the two isolated secondaries. Make a well damped tuned filter on the outputs.
From there use decent LDO's to make +/-13V That is more than sufficient for a 4V RMS signal through the TMUX'es
Not too expensive, and should isolate the circuit from the mains enough not to be measurable.
External medical grade "brick" for mains -> 24V DC. It can live outside the box.
And inside, a medical grade isolated 24V -> +/- 15VDC. Audio ground is made between the two isolated secondaries. Make a well damped tuned filter on the outputs.
From there use decent LDO's to make +/-13V That is more than sufficient for a 4V RMS signal through the TMUX'es
Not too expensive, and should isolate the circuit from the mains enough not to be measurable.
I have tested their medical grade supplies and confirmed with support that they still have a Y cap going from output to input for EMC reason. So if you are looking for true isolation, they are not it.
OP
- Joined
- Mar 7, 2020
- Messages
- 1,203
- Likes
- 1,917
- Thread Starter
- #90
The more I think about this part, the more I wonder if integration with ABX software on a PC is the way to go, with the software instructing the unit which input to select, and therefore also able to provide and fade/noise/whatever to both DUTs simultaneously. Effectively a hybrid of what @voodooless has been suggesting. Otherwise we need to generate it on-device, which is fine but less flexible.
Then again, is there a source for the gap requirement? I know I don't like having one when I use foobar ABX, and our switching time can be single digit milliseconds if we want.
True, but with the second TIM 3.5-2423 DC-DC convertet the total leakage would be in the sub 2uA range and probably less than the return currents that can be between the two DAC's. Probably a good idea to isolate the return paths in the DAC connections also.
There are many considerations in a product like this. Hats off to @mcdn for diving into this
OP
- Joined
- Mar 7, 2020
- Messages
- 1,203
- Likes
- 1,917
- Thread Starter
- #92
That's a really good point, I have a 250Wh one for camping. They usually have 12V DC outlets as well. I'll adjust the unit so it can run off 9-36V instead of 18-48, making it easy to use a 12V nominal battery for isolation.
This project badly needs a high level focus. Seems like we are designing before knowing that.
Let's say we build this. We throw to DACs or two amps at it and ABX results are positive. Then what? Does this mean all DACs/amps generate the same outcome? Or that just those two do. If the latter, then it is only useful for someone who is interested in that specific comparison.
A much better approach for amplifier is what the Swedish Hifi did with LTS. They took an amp, put a dummy load representative load, then using a voltage divider, brought its output down to that of the source. Then they fed both into another amplifier with AB switching against the pass through. In other words, we are listening for what colorations/distortions the amp added to the original source, a known quantity. Such coloration can be captured and analyze as well resulting in learning about the amplifier transfer function and reason for audibility.
The comparator was called F/E box (“F/E-lådan”). I asked ChatGPT Pro to create a schematic of it:
┌──────────── F path ─────────────┐
Source ── buffer/split ──┤ ├── F/E switch ── monitor amplifier
│ │
└─ amplifier under test ─ dummy load
│
precision attenuator
│
E path, level matched ─┘
The other issue is the goal for the outcome. If you are motivated to find no positive outcome you arrive at a different design than one that does everything in its power to tease out a result. The latter box for example would make it very easy to train, have features like controls, etc. I am in that camp. If we want negative outcome, then there is no reason to build anything....
Let's say we build this. We throw to DACs or two amps at it and ABX results are positive. Then what? Does this mean all DACs/amps generate the same outcome? Or that just those two do. If the latter, then it is only useful for someone who is interested in that specific comparison.
A much better approach for amplifier is what the Swedish Hifi did with LTS. They took an amp, put a dummy load representative load, then using a voltage divider, brought its output down to that of the source. Then they fed both into another amplifier with AB switching against the pass through. In other words, we are listening for what colorations/distortions the amp added to the original source, a known quantity. Such coloration can be captured and analyze as well resulting in learning about the amplifier transfer function and reason for audibility.
The comparator was called F/E box (“F/E-lådan”). I asked ChatGPT Pro to create a schematic of it:
┌──────────── F path ─────────────┐
Source ── buffer/split ──┤ ├── F/E switch ── monitor amplifier
│ │
└─ amplifier under test ─ dummy load
│
precision attenuator
│
E path, level matched ─┘
The other issue is the goal for the outcome. If you are motivated to find no positive outcome you arrive at a different design than one that does everything in its power to tease out a result. The latter box for example would make it very easy to train, have features like controls, etc. I am in that camp. If we want negative outcome, then there is no reason to build anything....
even the most perfect ABX box will face the exact same questions if the test result is from a singular person. multiple data points are essential.
imo this discussion is addressing the wrong problem. it can't be addressed with a box that singular people buy.
It can only be addressed with a source of accurate online recordings that are gained from a reference test setup, and inviting an audience to record their experience. which is what Archimago did. that's how you invite the kind of data pool that makes results credible.
imo this discussion is addressing the wrong problem. it can't be addressed with a box that singular people buy.
It can only be addressed with a source of accurate online recordings that are gained from a reference test setup, and inviting an audience to record their experience. which is what Archimago did. that's how you invite the kind of data pool that makes results credible.
The comparator “F/E-lådan” is a very diffenrent tool. It would be interesting to actually hear what kind of errors the amplifier does.
But it has its limits too. When testing with a linear impedance, you will miss how diffenrent output impedance will interact with a complex load. In other words, the test result may be identical, but with a speaker connected they can still sound different.
At which frecuency do you calibrate? A slightly tilted frequency response will quickly mask other errors.
But it has its limits too. When testing with a linear impedance, you will miss how diffenrent output impedance will interact with a complex load. In other words, the test result may be identical, but with a speaker connected they can still sound different.
At which frecuency do you calibrate? A slightly tilted frequency response will quickly mask other errors.
OP
- Joined
- Mar 7, 2020
- Messages
- 1,203
- Likes
- 1,917
- Thread Starter
- #97
Interesting. I hadn't mentioned it because it was lower priority, but this design does allow the use of the speaker terminals as the trim target for a source.
OP
- Joined
- Mar 7, 2020
- Messages
- 1,203
- Likes
- 1,917
- Thread Starter
- #98
That's absolutely valid, but a box that automates all the core parts of an ABX test and provides an audit trail does remove a lot of potential sources of error. It's also a very neat diagnostic tool for polarity inversion, channel imbalance, faulty devices etc. Not exactly a "one in every home" kind of thing, but the demand for the Van Alstine comparator shows there's some utility, and boy can we do a lot better than the Van Alstine! Technology's moved on, and even a "Better Van Alstine" is useful I believe.
OP
- Joined
- Mar 7, 2020
- Messages
- 1,203
- Likes
- 1,917
- Thread Starter
- #99
That's $50! But amazing specs. At least in the field of DC/DC converters it seems you do get what you pay for. Nobody here will accept anything less so I've put it in the preferred parts list.
OP
- Joined
- Mar 7, 2020
- Messages
- 1,203
- Likes
- 1,917
- Thread Starter
- #100
For better or worse, the focus at this time is "make Erin a state of the art ABX/switching device". Could end up a Homer car or could end up awesome, but if there's one thing a product needs it's customer #1. He does keep bringing up HDMI though, which makes me nervous so I'll stick to getting the analog part right for now.
Similar threads
