Now this is a story all about how,
My M4 got flipped turned upside down.
I'd like to take a minute, just sit right there,
I'll tell you how this Motu has inputs that aren't quite fair.
I picked up a Motu M4 with ESS ADC and DAC. It’s a great unit overall, but since I use it for measuring other gear, I wanted to see if I could improve the ADC side.
Most of my experiments didn’t lead anywhere, so let’s skip straight to the main issue: the DC bias on the line inputs is set incorrectly for the ESS ADC. This makes the inputs clip a few dB before reaching full scale.
The fix is simple if you’re comfortable soldering. The ADC actually requires a 1.95 V DC bias, but the Motu only applies 1.65 V. I understand why: the ESS datasheet says the bias should be half of AVCC (3.3 V / 2 = 1.65 V), which is common for many ADCs—but not for this one. ESS chips need 1.95 V.
What’s funny is that part of the M4 actually gets this right. The front XLR inputs on channels 1 & 2 are biased correctly at 1.95 V, but the rear TRS line inputs on channels 3 & 4 are at 1.65 V. It looks like two engineers designed different parts of the circuit, and only one of them knew the correct value.
Motu clearly noticed the clipping but didn’t know the cause, so instead of fixing the bias they just reduced the line input sensitivity. That way, the DAC outputs (about 4 V rms) can’t quite drive the inputs to full scale in loopback tests. I didn’t like that solution, so I first adjusted the line input gain to match the line output level, giving me a clean loopback without level changes. That only requires swapping four resistors per channel.
Now we can measure a loop-back with the input at -0.5dBFS and -1dBFS for comparison.
Now let’s look at the ADC bias. It’s generated by IC10, a dual op-amp. One half of the op-amp takes a divided-down voltage from the 3.3 V AVCC rail (the ADC’s clean analog supply) to produce the bias for the line inputs.
For reference, the AVCC rail itself comes from IC16 which I believe is an ADP151 regulator—it’s a small 3-pin SOT-23 device with about 9 µV noise. The package markings are unclear, but the pinout and performance match.
Here’s the important part: the other half of this same dual op-amp already generates the correct 1.95 V bias for the XLR inputs (channels 1 & 2). So the fix is simple—disconnect the divider feeding the line-input side, and instead link it to the correct 1.95 V bias from the XLR side.
(oops, that's not pin 8, it's pin 5)
Now let's run the tests again.
Nice
Now for some of the other experiments that didn’t work out:
I tried swapping the ES9840 ADC for the supposedly pin- and register-compatible ES9842PRO. In practice, it didn’t work. Most of the channels produced no signal at all, and channel 1 only gave a distorted, low-level output.
At first I thought I’d found the issue: the ES9842PRO datasheet lists AVCC as 4.5 V instead of 3.3 V. I tried supplying 4.5 V, but the result was the same—no usable signal.
It’s easy to assume I messed up the rework since the chip is tiny, but I checked carefully. The soldering was clean, all traces intact, and I even removed the chip, inspected everything, and tried again with a fresh IC. Same behavior. Swapping back to the original ES9840 restored normal operation immediately.
So, despite ESS calling them compatible, the ES9842PRO is not a drop-in replacement for the ES9840.
Next, I looked at the analog input op-amps. Motu use the THS2145, a differential op-amp designed for very low power and rail-to-rail operation—not exactly optimized for audio performance.
I swapped it for the OPA1632, which offers lower noise and distortion. At a -6 dB input level, the results were encouraging: I measured a noticeable reduction in distortion.
However, once I pushed the level higher, the OPA1632 swap revealed a problem. At about –3 dBFS the input started clipping. The issue seems to be that Motu run these input op-amps from a single +5 V rail (ground-referenced, no –5 V). With the input bias at only 1.95 V, the negative swing of the signal comes too close to ground, which most op-amps can’t handle gracefully.
Interestingly, the M4 does have a –5 V rail, and it’s used for the other op-amps in the circuit. I figured it would be straightforward to power the OPA1632 with that rail. Unfortunately, when I tied it in, the –5 V supply sagged to about –4 V and rose very slowly at startup. I guess that's why Motu stuck with a single-rail, low-power IC in the first place. A missed opportunity.
As a side note, I also replaced the AVCC regulator with an LT3045 (because I lost a leg on the original regulator). That did give me a measurable drop in mid- and low-frequency noise from the ADC, though I don’t have before-and-after plots to show.
My M4 got flipped turned upside down.
I'd like to take a minute, just sit right there,
I'll tell you how this Motu has inputs that aren't quite fair.
I picked up a Motu M4 with ESS ADC and DAC. It’s a great unit overall, but since I use it for measuring other gear, I wanted to see if I could improve the ADC side.
Most of my experiments didn’t lead anywhere, so let’s skip straight to the main issue: the DC bias on the line inputs is set incorrectly for the ESS ADC. This makes the inputs clip a few dB before reaching full scale.
The fix is simple if you’re comfortable soldering. The ADC actually requires a 1.95 V DC bias, but the Motu only applies 1.65 V. I understand why: the ESS datasheet says the bias should be half of AVCC (3.3 V / 2 = 1.65 V), which is common for many ADCs—but not for this one. ESS chips need 1.95 V.
What’s funny is that part of the M4 actually gets this right. The front XLR inputs on channels 1 & 2 are biased correctly at 1.95 V, but the rear TRS line inputs on channels 3 & 4 are at 1.65 V. It looks like two engineers designed different parts of the circuit, and only one of them knew the correct value.
Motu clearly noticed the clipping but didn’t know the cause, so instead of fixing the bias they just reduced the line input sensitivity. That way, the DAC outputs (about 4 V rms) can’t quite drive the inputs to full scale in loopback tests. I didn’t like that solution, so I first adjusted the line input gain to match the line output level, giving me a clean loopback without level changes. That only requires swapping four resistors per channel.
Now we can measure a loop-back with the input at -0.5dBFS and -1dBFS for comparison.
Now let’s look at the ADC bias. It’s generated by IC10, a dual op-amp. One half of the op-amp takes a divided-down voltage from the 3.3 V AVCC rail (the ADC’s clean analog supply) to produce the bias for the line inputs.
For reference, the AVCC rail itself comes from IC16 which I believe is an ADP151 regulator—it’s a small 3-pin SOT-23 device with about 9 µV noise. The package markings are unclear, but the pinout and performance match.
Here’s the important part: the other half of this same dual op-amp already generates the correct 1.95 V bias for the XLR inputs (channels 1 & 2). So the fix is simple—disconnect the divider feeding the line-input side, and instead link it to the correct 1.95 V bias from the XLR side.
(oops, that's not pin 8, it's pin 5)
Now let's run the tests again.
Nice
Now for some of the other experiments that didn’t work out:
I tried swapping the ES9840 ADC for the supposedly pin- and register-compatible ES9842PRO. In practice, it didn’t work. Most of the channels produced no signal at all, and channel 1 only gave a distorted, low-level output.
At first I thought I’d found the issue: the ES9842PRO datasheet lists AVCC as 4.5 V instead of 3.3 V. I tried supplying 4.5 V, but the result was the same—no usable signal.
It’s easy to assume I messed up the rework since the chip is tiny, but I checked carefully. The soldering was clean, all traces intact, and I even removed the chip, inspected everything, and tried again with a fresh IC. Same behavior. Swapping back to the original ES9840 restored normal operation immediately.
So, despite ESS calling them compatible, the ES9842PRO is not a drop-in replacement for the ES9840.
Next, I looked at the analog input op-amps. Motu use the THS2145, a differential op-amp designed for very low power and rail-to-rail operation—not exactly optimized for audio performance.
I swapped it for the OPA1632, which offers lower noise and distortion. At a -6 dB input level, the results were encouraging: I measured a noticeable reduction in distortion.
However, once I pushed the level higher, the OPA1632 swap revealed a problem. At about –3 dBFS the input started clipping. The issue seems to be that Motu run these input op-amps from a single +5 V rail (ground-referenced, no –5 V). With the input bias at only 1.95 V, the negative swing of the signal comes too close to ground, which most op-amps can’t handle gracefully.
Interestingly, the M4 does have a –5 V rail, and it’s used for the other op-amps in the circuit. I figured it would be straightforward to power the OPA1632 with that rail. Unfortunately, when I tied it in, the –5 V supply sagged to about –4 V and rose very slowly at startup. I guess that's why Motu stuck with a single-rail, low-power IC in the first place. A missed opportunity.
As a side note, I also replaced the AVCC regulator with an LT3045 (because I lost a leg on the original regulator). That did give me a measurable drop in mid- and low-frequency noise from the ADC, though I don’t have before-and-after plots to show.
Last edited:
