Michelle got me a Zoom H4 recorder for Christmas. It makes surprisingly good recordings for a portable, but I noticed a low level distortion, almost but not quite in the noise floor. So I measured its specifications to see if I could find it. Measurement showed a pulse peaking at -46 dB. The pulse consists of 700Hz and 250Hz components and pulses about 1.3 times per second when recording at 44.1 kHz. Other than that the overall S/N ratio seemed good for a portable - around -84 dB in the midrange and treble, decreasing (worsening) to around -60 dB at 20 Hz.

To be exact, the pulse was 689 Hz at 44.1 kHz and 750 Hz at 48 kHz. Notice anything strange about that? Divide it out and you get 64 in both cases. I suspected the Zoom H4 had a small buffer it was periodically writing out to the SD card. 64 samples at 16 bits each with 2 channels would be a 2048 bit or 256 byte buffer. Since writing to the SD card takes extra power, this could cause a transient voltage drop that the preamp might pick up as a pulse. If so, stiffening the power supply would eliminate or at least reduce the problem.

So I added a few capacitors to the H4's internal AA power supply to see what would happen.

Long story short, here are the results:
The original noise spectrum (not bad for a battery powered hand-held device):

The final noise spectrum (with 4 Kemet tantalum SMD 1000 uF 6 mOhm caps added):

In short, the pulse is about 12 dB quieter, the S/N ratio is improved by 12 dB in the treble and 3-6 dB in the bass.

Here's the long story:

Here are pictures of this initial surgery:
the unmodified Zoom H4 guts
The - battery terminal joins the board on the left, the + on the right.
Another view

Here are the parts. I ended up returning the capacitors on the left - they were too fat to fit. I used capacitors from the random assorted grab pack on the right. I figured since this is power supply regulation, nothing in the audio signal path, that Radio Shack electrolytics would be best. They give the highest capacitance for small physical size and for power regulation their poor ESL/ESR shouldn't be an issue.
Capacitors

First I added a 100uF cap. The smallest was too fat to fit into the space between the batteries, so I tucked it underneath, between the two circuit boards:
100uF cap, tucked in
100uF cap, wired into the other side
100uF cap, top (bottom) view

Unfortunately, upon subsequent testing this did not affect the pulse very much - reduced it only by about 1 dB (hey - at least it didn't get worse). But it did improve the overall S/N ratio by more than that. It was still around -84 dB in the treble, but now it remained at -84 all the way down to 100 Hz. Before, it started rising (getting worse) around 250 Hz. So I got a better S/N ratio from 20 Hz to 250 Hz, about 3-4 octaves. For example at 40 Hz S/N went from -69 to -78 which is a big improvement.

But the blip was still there... my next idea actually came from somebody else. Based on an internet discussion at the Zoom site, a chap suggested that perhaps the frequency components of this pulse - 250 Hz and 700 Hz - were too fast for the 100uF capacitor to respond. So I added one 47uF and one 22uF to test this theory. The idea was that the big capacitor would eliminate power supply ripple in the bass (which it already did) while the smaller ones would be able to respond at higher frequencies, extending the S/N gains into the midrange and treble, perhaps eliminating the pulse itself.

Well I put them in:
47uF and 22uF joined at the hip, but not yet installed
a little piece of tape to prevent electrical contact with the circuit board
All caps soldered on board

Did it work? Test signals would be the judge. A difference showed up in the FT analysis of the noise spectrum but it was so subtle I doubt I would hear it. With the dual caps the noise spectrum is smoother, slightly lower in a few areas, and nowhere higher. So overall, the smaller caps didn't hurt anything, but their improvement was marginal.

Not (yet) willing to give up, I orderd a few very special capacitors from Mouser Electronics. These are tiny, 1000uF tantalum SMD capacitors with a very low ESL/ESR of around 3-6 mOhm. I added three at a different location in the H4, in parallel with an existing 220uF capacitor that is on the output side of the DC-DC converter that the H4 uses when on battery power. I added one more to replace an existing 220uF cap nearby. This improved the S/N more effectively, but still not as much as I had hoped. It lowered the pulse by another 2 dB and the 690 Hz component in the FT noise spectrum by 12 dB. Overall the S/N is noticeably better than before, BUT the pulse is still there.

Here is what my H4 looks like inside now:
with Kemet SMD caps

In conclusion, I should mention that my test was simply to recording nothing while on the highest (+42db) gain. But this means the preamps are idling. If there were a real musical signal, the preamps would be drawing power and their power draw would vary dynamically depending on the signal they were amplifying. Since I stiffened the supply, it now provides a more consistent voltage under dynamic load. So my testing methodology tends to show the conservative worst-case improvement. It is likely that this mod also enabled the H4 to record transient dynamic signals with lower distortion.

The net result of this project is that:
1. I tested my initial guess (pulse being power related) and found it to be partially correct.
2. I learned a little about the importance of low ESL/ESR capacitors, even in power supply applications.
3. I have the satisfaction of having modified my H4 without sending it up in smoke.
4. I wish I had been able to eliminate the pulse entirely, but I can't complain - it's much better.

So I didn't achieve all I had hoped, but it was still a good result for a little experiment. So the $64k question is, why doesn't Zoom use better caps to begin with? One word: cost. The Zoom retails for around $300. It probably costs Zoom about $20 for the entire circuit board. The fancy caps I installed are about $10 each. $40 might seem a small price to pay for better sound quality, but it's almost certainly more than the entire board cost them to make.