New Isolator from TI

Is this post on EEVBLOG of any interest? Thought it may be useful.


Hi @djsb, yes, that’s a very interesting part for sure.

I’ve been using the ADUM4166 Eval Board for a few months now. It does a great job.

Hi Matt,
since the upcoming Q404 doesn’t have isolation, would be doable if I get one of the EV board proposed in this thread, and connect the Q404 USB to the computer THROUGH the EV board? This way the BNC shells will NOT be at the same potential as my PC, right?

Hi @Clane, yes, I think you are correct

Hi all
I’m not sure if this part really solves the problem. It allows to isolate the PC (USB host) from the QA4xx but you still need a clean power supply for the QA4xx. The power supply is definitely a challenge to achieve very high performance. I’m curious how this is solved in the QA404.

Hi Avo, typically the incoming power is run through an isolation transformer and rectified. That outputs an isolated DC. That DC is then turn through an LDO that has 60 dB of PSRR at the isolation xformer switching frequency. And then, the opamps have another 40 dB of PSRR at the isolation xformer switching frequency. And then the converters have something like 100 dB of rejection at the switching frequency. On the last page of the QA403 product brief (first link), you can see the ADC Aliasing test. In that test, and 346 kHz signal was injected to the input of the QA403. That produced an alias tone at 10 kHz. The input signal (at 0 dBV @ 346 kHz) was attenuated by the ADC more than 120 dB. And that same attenuation would be applied to the DCDC switching products. But overall, there’s probably 220 dB of attenuation to the DCDC switching frequency, which far more than needed. (That was for the QA403. The QA404 currently uses DCDC to convert the incoming voltage to the needed rails, but skips the LDOs for the +/-14V rails. The incoming 5V is also used to generate a +5/-5 (using DCDC boost for the 5V generation). That then runs through the 4.5/3.3V LDOs to power the ADC and DACs. The QA404 is also using very low-noise reference generation which the QA403 didn’t have.)

Now, in-band USB power disturbances can pose a bigger problem. DCDC converters take a few cycles to “see” there’s an incoming voltage disturbance and adjust the PWM duty cycle accordingly. This means that in-band power supply glitches will result in-band disturbances in the output of the DCDC. To combat that, you want a really fast DCDC switching frequency a lot of low-ESR capacitance after the DCDC converter so that you can “fight” the disturbance. And then that is followed by an LDO. And because the frequency is low, the PSRR is probably 80 dB or so. But the ADC can’t help here, because it’s all in-band.

Fortunately, if you have a very noisy USB power supply with really bad in-band noise, you’ll see it clearly in the spectrum. But I’ve not seen a supply yet that is bad enough. The second link shows a QA401 running directly from the USB of a fanless PC without any impact. USB powered hubs are generally cleaner still.

I have found that testing DACs can be a challenge. Not many of these products go to the extent that QA has gone to in order to isolate from common mode currents through the USB connection.

So, as I mentioned before, I have been using an ADUM4166 eval board to connect to the DACs. The eval board is powered from a linear wall wart power supply from Jameco, which then ultimately powers the VBus line for the DAC. So far, that remedy has been completely successful and lots of artifacts are removed from the spectrum.

Side note - There’s some well known web sites that publish test results for DACs and the rest generated from an Audio Precision test system. As good as the AP is, with it’s special power transformer for maximum isolation and careful execution of transformer isolated analog inputs and outputs, you can often see artifacts in the spectrum that probably are caused by common mode currents. These are often either ignored or pounced upon by the testers, depending on I’m not sure what. Rarely are they identified for what they are. I guess there’s a lot of morals you can take away from this, but a big one is that testing audio equipment in isolation in an ideal operating environment isn’t really a complete test.

I know this thread (Topping/Cosmos Based Test System) isn’t about QA products, but the author (not me!) does show good examples of what happens without proper isolation.

Hello Matt
Thank you this very good explanation of your power supply approach. For my pre amp design, I use a 115V/240Vac to 12Vdc converter followed by a switching power regulater in SEPIC topology for creating split voltages and low noise, high PSRR LDOs to remove high and low frequency ripple. Some high-enders may claim switching power supplies are not good for high qualiy audio but I see just pros and no cons.
For my purpose AD have very good switching regulators and LDOs. Part availability seems better than TI but as of now I only need low quantities for pre production. I’m looking forward for the availability of QA461 and QA404

Hi @BKDad, the USB isolation is a big issue to try and untangle for sure. I will need to check, but there is progress on getting isolation back into the QA404. In the QA403, there are three isolation transformers: One for 3.3V, one for 5V and one for +/-14V split rail. But the transformers drivers (SN6505B) are very hard to come by right now. And so, a new approach that uses just a single SN6505B and a larger transformer is being tried as some SN6505B have recently arrived. Hopefully it shows some promise.

the plot below is REVC version of QA404 being driven by a Topping D10S in balanced mode. Both are connected to the same PC. This version doesn’t have isolation. There’s some fussing about to deliver the best performance (eg D10S frequency, D10S level). Note the max input on the QA404 is shown as +12 dBV, but it’s slightly over-driven (12.13 dBV). This is because there is about 0.5 to 1 dB of margin left to account for unit to unit calibration. That is, in 12 dBV max input setting, the max input will be 12.5 dB on some units, and 13 dBV on other units.

This also isn’t using any tricks to help with noise (eg mono mode) or harmonic cancellation. The mono mode would help here, as the THDN is limited by noise. And once in mono mode, THEN the harmonic cancellation would help

But, in summary, a good FDA (like OPA1632) will do a lot of the heavy lifting to knock-down the CMRR noise. I’m not certain, but I think the Cosmos E1DA closely follows the ESS reference design. In that reference design, there is just a pair of OPA1612 opamps going into the ADC, and the ADC is then responsible for cancelling out the common mode noise. However, the ADC doesn’t have a spec for the common mode rejection. And so, it’s really important to have a super-wide bandwidth FDA with lots of open-loop gain to knock-down the common mode BEFORE it hits the ADC. In other words, don’t let the noise into the ADC at all. Kill it in the FDA with bandwidth :wink: I think the plot here shows how well it works.

That said, I really hope the isolation can make it back into the QA404. More experiments are under way.

PS. The plot also has +/-100V protection in place on the inputs. This is new on the QA404, and it means that even in the most sensitive 0 dBV mode, you can touch the inputs to any voltage from -100V to +100V without issue (with the atten on the protection extends far beyond +/-100V). The same is true for the outputs. There will be more info on that later. But the aim is to continue making the the IO bulletproof against probe slips. All subject to change, of course.

I have one of the Cosmos ADC’s here. I’ve never taken it apart to reverse engineer it, so I can’t say whether it’s the same as the ESS Reference design. Besides, I don’t know what the ESS design looks like anyway!

But, I can tell you that I am getting similar results to what you showed with a D10s, a Jan Didden Autoranger, the Cosmos ADC, and an ADUM4166 isolator. Certainly, what I measure is within reasonable unit to unit variability. The two DAC channels are slightly different from each other, for example.

Plus, not to rub it in, I was able to purchase all of these. Good timing on my part, I guess. Luck counts for a lot. They get used along with the QA480 and QA401 in my hobbyist “lab.”

Without the ADUM isolator, the spectrum is kind of ugly. Apart from the implications in a test system, it makes me wonder what things are like in an actual home hifi system, where the CMRR of the preamps and power amplifiers are probably not so great. Especially with unbalanced interconnections.

BTW, isn’t the OPA1632 sensitive to source impedance mismatch with regard to CMRR? I know there’s ways around that, but everything adds up in terms of noise, distortion, cost, power, pcb space, etc.

Yes, but I guess the point is that front-end CMRR works WITH the ESS CMRR. I point this out because the belief at the link you shared was that “This is bad because there’s no isolation” whereas I think the correct response is “If CMRR isn’t considered, then isolation is required to eliminate common mode noise.”

Also, the CMRR problems you note with FDA amps would also manifest in a differential ADC, right?. If your signals are imbalanced due to impedance mismatches, resistor gain mismatches, etc, you are stuck whether it’s being done in the analog or digital domain. Because the math is pretty simple.

I don’t have the definitive answer here. I’m just stating the notion that isolation is required to eliminate ground loops in differential paths fails to acknowledge what decades of differential signaling have accomplished. Isolation is hugely important in single-ended measurements, however. it’s been in every QA analyzer since the QA401, and I really hope it can make it into the QA404.

I can’t argue with your comments, because I agree. So, I won’t.

Couple points:

The CMRR of most instrumentation amps tends to degrade with frequency, as you’d expect. So, common mode junk outside the audio band will still cause problems with most circuits. The source mismatch only makes it worse. I think this is why additional isolation is worthwhile, because, as you say, it all adds up.

TI has a suggestion in a note they publish about dealing with impedance mismatch. It’s not really original from them, but is easily accessed and might be of interest if you haven’t seen it:

TI on CMRR - Make Rcom over 1 Meg.

In a practical world, things are never like they are in simulation, You get common mode to differential mode conversion in cables, for example. These all limit the floor of what you’re trying to measure. Best to use a belt, suspenders, and whatever else you can manage. Degradations add up like isolation does.

The other point I’d add is that audio system components used in real systems are exactly as you describe about the differential ADC. That is, imperfect CMRR for a host of reasons. Perhaps you might wish to consider adding provisions in the test software to make a CMRR test easy and calibrated. Maybe nobody will care, but maybe you’ll help point to a problem that is often ignored.

My soapbox comment is that as you and the incumbent test system guys make the measurement gear better and better, you actually make the measurements less revealing in a sense. Having perfect CMRR at the input and output of the test instruments and power source isolation as high as possible certainly reduces the “Heisenberg” effects of the test gear. The other side of that card is that the test gear also doesn’t reveal DUT performance in non-ideal operating conditions. That may be a matter of educating the test system users, which might fall out of your company charter. But, it’s still a thing in a world where lots of people judge the quality of a piece of home audio gear by tenths of a dB differences in SINAD and where they also judge the quality of a power amplifier for its ability to drive a purely resistive load through no cable.

It’s a pretty easy measurement to make. On an opamp, you might do the following:

In the left schematic, you drive the opamp terminals with the exact same signal. The gain here is unity, so it might be hard to discern the opamp CMRR from the analyzer CMRR. The second circuit on the right has 60 dB of gain, so it’s easy to force the opamp to reveal its CMRR. Input a 0 dBV signal, and if the opamp has 120 dB of CMRR, then the 0 dBV signal shows up as 60 dB.

On an analyzer, you’d do the following:

And then drive with a 0 dBV signal and then measure the resulting level. You will have 8 different readings, because each of the 8 inputs will have its own CMRR that is determined by the performance of the 0.1% thin film resistors used in the front front end. On the particular QA403 on my desk, the best CMRR is attained at +12 dBV. Here we can see the 0 dBV signals applied to the + and - inputs are attenuated -95 dB.

The CMRR for the other full scale inputs are as follows:

0 dBV Max input -91.8 dB CMRR
6 -79.61
12 -94.9
18 -72.2
24 -81.3
30 -76.9
36 -82.1
40 -74.2

Now, as you note, you can see the CMRR varies quite a bit. And there are calculators on-line to help you figure the minumum and typical CMRR given the gain and resistor tolerance. For example, with 0.1% resistors, an FDA with unity gain will see a minimum CMRR of 54 dB (76 dB typ). And 0.01% resistor would yield 74 dB minimum (96 dB typ). In a world of unlimited budget, you’d hand-tune a pot in the front-end to drive down the CMRR to some satisfying number. But once you add pots to the design, then you get into the treadmill of annual calibrations which has its own drawbacks.

But rather than treating variation as a problem, it can be a benefit too. For example, on the analyzer noted above, the CMRR difference between the 6 dBV full scale input and the 18 dBV full scale input is 22 dB (this will vary unit to unit). But this means that simply switching the input from, say, 12 dBV to 18 dBV will result in a shift in CMRR performance. And if your signal is showing levels of hash changing as you move from 12 to 18 dBV, you can know it’s due to CMRR.

OK, but what about times when you really, really need killer CMRR performance? In the post HERE there is discussion of a Line Receiver (pasted below):


This is a special circuit that is focused on CMRR. Note in the front-end the 0.1%. But take a look at R19 between the two OPA1612 opamps. Like your earlier post on how to reduce the impact of imbalances, the schematic borrows from the TI INA1651 line receiver:

In the INA1651 spec, TI goes through the math for the different center V values:

The tradeoff is settling time for the circuit.

The backend of the circuit performs a balanced to single-ended conversion, The resistor array on the right side of the schematic is 0.1% absolute, but matching is +/- 5 ppm (0.0005%). At that level, the CMRR is more likely an issue with routing.

Probably more info that you wanted, but hopefully interesting to some.

Thanks for the detailed information. That’s pretty much how I’ve made CMRR measurements up until now. It’s good to have some validation for that.

So, I just made a couple tests.

The first is of my QA480 oscillator into the Autoranger and the Cosmos ADC. Red is with no USB isolator and Black is with an ADUM4166 isolator in the USB line to the QA480. I did my best not to move any cabling or anything like that. Just add in the isolator. As shown, there is some improvement in distortion with isolator, but not a lot.

The second is of a Topping D10s. Same color code for the plots. The isolator offers more than some improvement.

I guess this shows the limitations of an ADC that has 50-55 dB of CMRR when there is a complete ground loop.

Earlier on, when I was debugging all of this, I tried connecting the ADC to a different computer in the same room, also running REW. Everything else was the same. The results were similar to what is shown for the D10s with no isolator. Yeah, it was different, presumably because the common mode current loop was now through the AC mains and the power supplies of the two computers instead of across the USB system in a single computer. I also tried using the ADC with a laptop running on battery power as the spectrum analyzer. It was essentially the same as the single computer operating when an isolator is in place to break the current loop.

So, I guess it pays to watch for common mode currents. They are probably there, and might cause overall system problems even if the test system doesn’t show them. It also pays to have as much isolation as you can manage without causing unwanted side effects.

I’d guess that 50 dB CMRR might be the mean value for home audio products. Some very well done balanced systems have better CMRR. Many not as well done balanced systems are around 50 dB. Unbalanced systems, well…

I find all this information very interesting.Thank you

1 Like

Hi Matt
A new chip, the SN6507, seems available from TI. It has just a 0.5A switch but is more flexible compared to thr SN6505. The chips are not pin compatible. I guess you know about this part. Are there any particular issues with this chip or do you see it useful for audio power supplies? I think with 2 external FETs the limited drive capability can be increased to much higher values. This may have drawbacks to achieve highest efficiency but efficiency is at least for me not the most important design criteria.
Availability of the SN6505B seems still a problem until 2024 and beyond (Digikey and Mouser have up to 120!!! weeks lead time)

Hi @Avo, yes, that’s a good part too. The higher voltage makes it much more suitable for a 24V incoming supply (for example), which gets split down to +/-15V. At the USB lower limits of 4.5V or so, the 0.5A limit leaves you with about 2W of total iso power which is a bit tight.

What is going on with TI?

Here’s TI’s web site:


Mouser has them in stock, but DigiKey lists them as obsolete. Who knows? Good luck in finding an eval board, too.

But, ADI has some similar parts, mostly from their various acquisitions over the past few years. I’m going to investigate those.

Tough to design around, for certain.