Improvement of the ESI U24 XL USB sound card

Is it possible to improve the performance?

Abstract

When I revised the review of the ESI U24 XL USB sound card, I noticed some potential issues and started to think about improvement of the sound card. I decided to collect my ideas in this article instead of the review to avoid cluttering the review too much. Swapping components inside audio equipment is not too uncommon. This is mostly done during restoration of old equipment because some components like electrolytic capacitors have a limited life expectancy. Swapping operational amplifiers is also pretty common and the rationale behind is to improve audio performance. This may or may not work well. I haven't modified any equipment previously and found this to be an interesting project.

Analog Input

Analog Input Schematic

Analog Input Review

The analog input circuit puzzles me a bit. I believe that capacitors C28 and C31 should better be located after R50 and R61, i.e. in parallel with R53 and R58 to form a well defined low pass filter keeping high frequency disturbance away from the analog input. Considering the rather low slew rate of the NJM4580 op-amp, the chance to precisely limit the slew rate at the op-amp input is missed. Also, the cable inductance may resonate with this capacitance. I guess that capacitors C28 and C31 were placed this way deliberately in order to ensure EMI compliance. It took me a while to get the point of those capacitors location: The power supply generates so much high frequency disturbance that the shielded enclosure is there to protect the outside world from radiation within and this is probably why they designed the filter the other way round being a low pass towards the cable. The filter is simply not meant to keep high frequency trash out, but to keep it inside. It hurt when I finally realized that.

The op-amps are run from asymmetric supply voltages (+5V and -3V). Any DC offset however is blocked by capacitors C42 and C47 so DC offset at the op-amp outputs does not present an issue to the ADC. I find asymmetric supply voltages a bit inelegant.

Analog Input Improvement

The NJM4580 is a budget op-amp (0.5€ when purchased in low volume), but not a bad choice and there are worse options on the market. Considering the performance of the ADC and all the noise generated by the power supply, upgrading the op-amp might not yield serious improvement. Nonetheless, I might try that one day just out of curiosity because I haven't upgraded any op-amp before and like to see whether it could improve anything. Stability of a different op-amp in a given application is another story, but C45 and C46 (plus R51 and R59) should help maintaining stability.

Potentially suitable upgrade op-amps that maybe fit the SSOP8 footprint of the PCB could be the OPA1692 (1.7€) or OPA1602 (2.5€). The latter one would be my preference. I haven't further investigated fitness of the potential upgrade op-amps for this purpose and merely put them on my watch list in case I ever like to experiment with an op-amp upgrade. Both have packages that are smaller (MSOP / VSSOP) and may require some solder work of art in order to fit.

Analog Output

Analog Output Schematic

Analog Output Review

The analog output is connected right to the DAC pins 26 and 27. It is sad to see that ESI did not follow Wolfson's recommendation to implement a low pass filter ahead of the op-amp (see figure 44 of the data sheet). A low pass filter would have cost a few cents and there is still plenty of PCB real estate to implement such a filter. Well, Wolfson recommends a low pass filter for HiFi applications. Maybe ESI did not target HiFi during design of the sound card.

Capacitors C54 and C55 seem serve the same purpose like C28 and C31, which is to ensure EMI compliance is met. Putting a capacitor at any amplifier output degrades stability of this amplifier. Here, the 100Ω resistors help to isolate the capacitance and maintain stability. Again, I find it a bit inelegant to connect a capacitor directly to a wire since this may resonate and in this case the capacitor is also parallel to the headphone. Splitting R22 and R23 in two with each half the resistance value and putting C54 and C55 in between the two resistors, forming a "T" filter, would be better in my opinion because this would also form a more effective filter to keep EMI away from the amplifiers output, which is also the inverting input. Omitting C53 and C54 entirely could be an option.

Analog Output Noise
ESI U24 XL analog output noise 10mV/div 10µs/div
This is what the noise at the analog output looks like with the oscilloscope set to 10mV/div and 10µs/div. The probe was set to 1:1 so there is some input and cable capacitance that contributes to forming a low pass filter. The capacitance may be up to 100pF roughly. Honestly I don't remember exactly whether this picture was taken prior to removal of C54 and C55, but as pointed out, the scope provides some capacitance anyway. The filter ESI built would roll-off at 7MHz and the picture does not show anything meaningful above 1MHz. Note that this is the noise with the output unloaded. In case a low impedance load like a headphone is connected, the noise is divided down due to the high output impedance of the analog output. The noise stems from the DAC so the op-amp is not to blame. DACs output extreme levels of high frequency noise at the output so this is not a big surprise. Lowering the DAC noise maybe could be accomplished by lowering the analog power supply noise, which would also benefit the whole analog output and input section.
Analog Output Clipping
ESI U24 XL analog output clipping
Just like the analog input, the output op-amp is run from asymmetric supply voltages (+5V and -3V). This means that DC blocking capacitors C9 and C12 are mandatory. And this also means that clipping will occur in an asymmetric way, i.e. the negative clipping will occur much earlier than positive clipping. The good news is that recovery from clipping does not look bad. The picture shows a 1200Hz sine wave being clipped. Note that this picture was taken with the output unloaded.
Analog Output Frequency Response

Output DC blocking capacitors C9 and C12 do not have enough capacitance to properly reproduce low frequency signals into low impedance loads. With 32Ω headphones, -3dB attenuation would happen at 26Hz and some lower attenuation as well as phase shift occurs much higher in the frequency range. This is a pretty obvious design flaw. 47µF capacitance are okay with a 10kΩ line input, although this might already cause some phase shift in case the line input has the same capacitance for blocking DC installed because both capacitors are in series and the total capacitance is reduced dramatically.

Here is a frequency and phase response plot of the headphone output unloaded and loaded with a 43Ω resistor.

Analog Output Improvement

Increasing the capacitance of C9 and C12 dramatically improves bass response with low impedance headphones. I did some research and think it is possible to fit 330µF instead by increasing the capacitor size from 5mm to 6.3mm, which should still fit on the PCB and pattern. This results in roll-off at roughly 4Hz with 32Ω headphones and is somewhat acceptable.

A possible upgrade for the op-amp could be the OPA1688. When using this op-amp, the output resistors R22 and R23 should be lowered to maybe 50Ω and the capacitors C54 and C55 should be removed. I guess this would be an audible change, especially with low impedance headphones since the amplifiers output impedance is much lower after the upgrade. The OPA1688 is not available in DIP package so an adaptor would need to be installed.

Power Supply

Power Supply Schematic

Power Supply Review

Reliable and quiet power supplies are the foundation for every circuit. Components required to achieve a good power supply are usually expensive and I would not be surprised in case the power supplies performance was subpar.

I measured ripple on the +5V and -3V supplies and was not surprised to see 80mV peak to peak triangular shaped ripple on the +5V supply and roughly 30mV high frequency noise plus some ripple on the -3V supply. Both are used as analog power supplies.

The +3V3 supply for the digital part is derived from the USB power using a 1117 series regulator. While there is adequate capacitance at the linear regulator, the input capacitor is right next to the input capacitor of the boost converter and therefore is not dedicated to the linear regulator, but actually belongs to the boost converter instead. All the boost converter noise is present at the linear regulator input as well. However, the linear regulator does a good job suppressing the noise at its input and providing adequate output voltage for the digital circuitry.

For the digital part there is also a +2V5 and +1V8 supply. The +2V5 and +1V8 supplies are generated by the ICs that require those supplies.

The +5V and -3V supplies are generated by a boost converter that uses the USB power as input. The negative voltage is generated by a charge pump that is fed by the boost converter switch node. The inscription on top of the IC was not helpful identifying the IC. It seems to have a SON 10 package likely with exposed pad underneath.

This is how the noise on the power supplies looks like:

Positive analog supply 10mV/div 10µs/div

Negative analog supply 5mV/div 10µs/div

From the ripple wave form I would conclude that the boost converter runs in some kind of burst mode. A short group of switching pulses is followed by a long pause and once the output voltage decreases to the lower threshold after roughly 30µs, another group of switching pulses ramps up the output voltage to the upper limit. This is best visible on the negative supply. The boost converter seems to run at roughly 2MHz switching frequency.

Here are some oscilloscope shots of the switch node and +5V output voltage AC filtered. The first picture illustrates the burst mode operation at 5µs/div. The second one shows more detail at 1µs/div and the third one at 500ns/div shows even more detail of the ringing associated with the switching of the converter.

Using a charge pump to generate the negative supply voltage is super cheap, but also pretty inelegant. The designers from ESI had better read the article from Texas Instruments about charge pumps explaining how to do that properly and may have learned that a converters running in burst mode are not the best foundation for such a solution. Since ESI also omitted the resistor that limits loading of the switch node by the charge pump, this could be an explanation for the ill behavior of their power supply I observed during my investigation.

Power Supply Improvement

Improving the power supplies is more difficult than it seems. Usually throwing large capacitance at the problem is a good cure, but there is a switch mode regulator involved, which either may go unstable or in over-current protection in case too large capacitance is placed at its output. Also, there are some ferrite beads installed for filtering noise and those have unknown current capability. Too much capacitance without current limiting factors may fuse the chip ferrites. Putting higher capacitance at the input of any switch mode converter is usually a good idea, but this runs from the USB, and the USB power controller of the PC may not tolerate too much capacitance connected.

I experimented a bit with the power supplies. Adding each 1500µF to C58 and C39 increased the negative voltage to -3V5. The noise floor of the spectrum now showed some weird humps and the audio signal inter-modulated with something, which formed quite a lot of extra frequencies in the audio spectrum. Clearly, this modification caused power supply issues by messing up the boost converter.

In another experiment I tried loading the +5V output of the boost converter using a 100Ω resistor. The idea behind is that a small extra and constant load, 50mA in this case, could bring the regulator in continuous conduction mode and elevate the negative supply a bit. This almost worked for a short time and the negative supply was increased to -4V, but the regulator totally freaked out shortly after. What would happen in case the circuitry attached, like the headphone driver, draws a bit more current? Would the boost converter start to oscillate then as well? Of course, yes, driving a slightly clipping signal into the 43Ω load with one channel causes some kind of oscillation that seems to affect the burst frequency.

The boost converter also does not seem to like a large ceramic or electrolytic capacitor at its input, which is unusual because according to my experience switching converters like a lot of capacitance at the input, especially low ESR and ESL ones like ceramics.

I installed a small 10Ω and 68pF snubber at the switch node to ground. This slightly reduced ringing, but also polluted the ground a bit. In order to be effective, the snubber would need to be connected to places that are not easily accessible. The ringing probably reaches far into the 100MHz range and probably beyond and it would be nice to attenuate this a bit. To some part this is an inherent problem of boost regulators.

If I only had an idea how to stabilize the boost converter. There are some unplaced components associated with the boost regulator, but all of them are resistors. Seems like the compensation is internal and therefore not configurable. And it seems that the application built around the boost converter is not really well engineered - else it would be stable. The converter seems to have a very high switching frequency of roughly 2MHz and oscillation seems sub-harmonic affecting the burst frequency. This is difficult to debug with only an old analog 100MHz oscilloscope.

Here is what I believe the other pins of the boost converter do:
Pin one is likely an enable or soft start control. I assume that once Q8 conducts, the converter shuts down. In case soft start would be controlled here, this would be implemented by adding a capacitor to ground.
Pin seven is likely some under-voltage lockout, which is unused due to the resistors being uninstalled.
I have no idea yet what pin four and R29 could do.
I was not able to identify where the pins missing from the schematic are connected. Probably those pins are unconnected.

Probably it is best to leave the boost converter alone and improve the filter after the converter. The chip ferrites FB2 and FB3 neither filter ripple or noise significantly, nor do they isolate the converters output from the rest of the PDN. This is no surprise because the capacitors C39 and C58 have way too high ESL to be effective at high frequency. A good instruction how to approach such a filter is an article about designing second stage output filters for switching power supplies from Analog Devices. This paper also explains why ferrites are a poor choice for such a filter.
Using 0805 size chip inductors instead of ferrites with maybe 4.7µH and 0.3Ω series resistance are good up to 500mA of current and will improve both ripple and noise filtering and is not resonant due to the high series resistance of the inductor. Adding series resistance to the filter inductor is another way of avoiding resonance in L-C filters not mentioned in the paper from Analog Devices. I find this a good method in case some voltage drop is acceptable. This now should allow to increase the capacitance of C39 and C58 above 100µF without affecting the boost converter resulting in a well behaved filter with Q=0.7 and roll-off at 7kHz. This looks extremely promising and is what I'm going to try next.

I tested above theory and changed FB3 for a 100µH inductor I found in the scrap and this reduced ripple to around 5mV, but the boost converter showed signs of instability again. I'm somewhat optimistic that the solution I have in mind could actually work, but this may require some experimentation to optimize the filter. I also experimented with improving the filter using different chip ferrites scrapped from another PCB. This improved high frequency attenuation a tiny bit in the high frequency range, but is actually pointless.

PCB Design

During reverse engineering, I became familiar with the PCB design of the sound card a bit. There is some segregation of the ground between analog and digital, which is not a bad idea. But this segregation does not seem to be well engineered since a lot of signals cross the segregation anywhere but at the ground merge point. This is a common problem with ground segregation: It is either not well thought out, not implemented properly on the PCB or both.

Conclusion

For now, the conclusion is that there is a lot of potential for improvement. When I started the investigation, I thought about upgrading components like op-amps. The longer I investigated the sound card, the more I realized that the design is not worth any upgrade, but needs to be fixed because it is broken by design. The power supply is terrible and therefore, the supply voltages are noisy and any improvement would be a huge advantage. Improving the analog output seems far more rewarding than doing anything with the analog input. Here is a ranking of which improvements I see as most rewarding:

  1. Increase C9 and C12 to 330µF. Cost: Roughly 0,90€ to 1,40€. This by far yields the most dramatic improvement correcting the frequency response of the headphone output and costs next to nothing.
  2. Lower ripple and noise on the power supplies. This is difficult and I'm still investigating how to best improve the supplies. No further improvement makes any sense unless the power supplies are fixed.
  3. Add a low pass filter in between the DAC and the headphone op-amp. Cost: Maybe 1€. This will filter high frequency noise and avoid inter-modulation of such noise with the audio signal. This might lower the pretty high noise level on the analog output that stems from the ADC.
  4. Upgrade the NJM4556 to OPA1688, remove C53 and C54 and decrease R22 and R23 to 50Ω. Cost: roughly 4€. This may yield a subtle improvement for low impedance headphone users.
  5. Upgrade the signal processing NJM5580 to OPA1602. Cost: 2,50€. I'm not sure whether this would really improve anything.

Modification

Here is how I actually modified the sound card:

Analog Output

First, I removed C54 and C55. I measured them as 220pF. I would say removal was unnecessary and such a low capacitance does little harm when placed behind 100Ω resistors. However, just like with the analog input, the capacitors are probably there to contain all the high frequency emissions generated by the power supply inside the enclosure.

I thought about lowering the resistance value of R22 and R23, but decided to keep them the way they are now.

I found some 1000µF / 10V capacitors at home and could manage to install them as C9 and C12 on the PCB. The bigger, the better - so why not?

This is how the frequency response looks like now loaded with 43Ω again:

As expected, low frequency roll-off is perfect now. Removing the small capacitors has changed high frequency phase a bit into the other direction, but now the phase is off by a third of the previous value only. Needless to say that the sound has improved dramatically because the bass is much better. THD and spectrum has not changed significantly.

Analog Output Filter

ESI U24 XL analog output filters comparison
It is near impossible to implement the second order low pass filter recommended by Wolfson on the board that ESI designed. However, I saw a chance to place a simple first order low pass filter ahead of the input of the analog output operational amplifier. Therefore, I removed C5 and C6 and improvised the filter at this place instead. I reused some components from another assembly to build this filter. The most suitable components I found were 470Ω resistors and 1n5 capacitors. This forms a low pass filter with 220kHz roll-off frequency and only small attenuation of the audio signal. Of course I also installed some DC blocking capacitors again. The filter is not optimal, but it reduces high frequency trash fed into the op-amp while neither affecting level nor phase in the audio frequency range too much. Feeding tons of HF noise into an amplifier is just a bad idea because the noise would intermodulate with the audio signal and the noise also exceeds the slew rate of any op-amp. Therefore, a passive filter is a very good idea and helps to reduce HF noise level a lot.

Analog Output Gain

I changed resistors R13 and R14 from 13k to 4k3 in order to increase the gain of the output amplifier. Note: The output line level complies to the usual -10dBV found in consumer equipment. Reson for the modification was that I wondered about low output level because forgot to increase software volume and fixed this in hardware instead. Stupid mistake, not a recommended modification. However, the increased gain of the amplifiers (over-) compensates the level attenuation of the filter I added ahead of the op-amps.

Power Supply

Every attempt to improve the power supplies has failed so far, but I have a plan how to fix this issue.