Digital source for frugal-philes

I recently obtained this SD card player:



For the money (I paid about $120 here in China) its a bargain. Plays FLAC from two SD cards, has digital and analog out (balanced too) as well as headphone amp. Build quality is well above average for Chinese made stuff.

The only drawback on the digital out is it has an in-built sample rate converter, upsamples 44k1 to 48k or 96k or 192k, so its not bit-perfect. But on the other hand, it means you could use an external NOS DAC and not get much droop

Analog outputs come from a CS4398 but straight out of the box didn't sound too great. I've done some modifications and improved it considerably, will talk about the mods in more detail on this thread if anyone is interested.

You just lost your bet Here's an internal shot to get you drooling:



The mods involve changing the output opamps - just above the big red fashion-caps. They supply NE5532 but they're socketed so easy to upgrade. I've gone for LM6172. But in isolation this doesn't help much - rewiring the PCB makes a big difference too. I cleaned up the PSU circuitry as well. I'll post up some shots in due course.

Some of the rationale behind the mods first - enemy no.1 for good digital sound is noise. By this noise I don't mean the kind of noise found as the 'N' in single tone THD+N measurements. That's static noise - I reckon if you measure the static noise of my modded player vs the unmodded it won't look too different. I don't have the kit to do that though so this is just conjecture. The problem is noise modulation - that is noise which changes dynamically with the signal. When there's no signal, there's no noise.

One of the ways the noise gets into the outputs is via non-linearity of semiconductors, particularly ICs. Opamps have a bad rap in audiophile circles and I think its primarily for this reason. They're susceptible to high frequency (meaning above audio band) noise, and can 'fold down' this noise (aka intermodulate it) into the audio band. One example of this is what's commonly called 'sibilance' - bipolar opamps do tend to generate this in high degree in digital systems. NE5532 is susceptible to this but its by far from being the worst offender IME - the newer, lower noise devices which have much better specifications (LME49xxx) can be worse in practice at creating sibilance. For this reason I tend to avoid using low noise opamps, or if I have to use them I need to protect them extremely well from HF hash. One not-so-low noise opamp which works well (SQ-wise) in harsh environments (digital) is LM6172. Its cheap and gives great sound for the buck, but its a little tricky to get stable - i.e. it oscillates with little provocation.

Here's a shot of the analog output part of the player, with mods implemented.

Firstly you'll probably note the fashion caps are out - that's because the output offset of the new stage measures less than 20mV and the best cap is no cap. Second the opamp sockets aren't populated - that's because I am using SMT parts which I soldered in on the reverse side of the board - pic to be shown later. Most of the passive components around the output filter/buffers have been extracted and some 0805 precision resistors (green in colour) replace them - hard to see unless you magnify the pic. There's a new output socket fitted - I use SATA connectors and cables because they're convenient, balanced and stereo. The connector is the shiny block just in front off the yellow relays which are used to mute the outputs.

Underside mods part 1



As promised, here's a shot of the underside where the power supplies are - the area to the left of the pic in the previous post where you'll note big caps and heatsinked regulators. There are three separate power supplies - two analog (positive and negative) and one digital. They're using 3 pin fixed voltage regulators which have poor rejection of high frequencies. So I give the regs a little assistance by installing passive RC filters prior to them. I split the original lumped reservoir capacitance into two and the second cap is preceded by some SMT resistors which are visible in this shot. SMT ceramic caps have better HF characteristics than large electrolytics.

You'll also notice the yellow wires converging on a point - this is star earthing. A major way noise gets introduced into the analog circuits is via common-impedance ground coupling. Put another way - noisy grounds. Plenty of designers go for a big ground fill because they think the more copper in play, the better. But then they ask their ground fill to do double duty - work as a sink for power supply noise (decoupling) and as a nice quiet signal reference. This doublethink just doesn't work in practice so I hack the ground fill connections with a sharp craft knife and replace them with explicit point-to-point wiring. The result is quite audible in improved separation of instruments in the soundstage as well as the soundstage itself growing in depth and often, width.

Another minor tweak is visible to the right - I added series resistors to the caps paralleled with the soft recovery diodes. I prefer to soak up RF noise rather than merely recirculate it The digital supply is the bottom one - this originally came with standard (i.e. noisy) diodes which I changed over to schottky type.

Underside mods part2

I've soldered on two LM6172s (upside down to keep the pins in the right positions) configured as standard differential amps. Since the CS4398 has differential outputs, two diff amps nicely creates the two complementary output phases for balanced drive, and also presents equal loading on the DAC's outputs. Note also that the balanced outputs (3 pin XLRs) grounds have the very common 'pin 1 problem' so the yellow wires to them are to correct that error.

JoeyGS said:

Have you considered using Burson HD Opamps (http://bursonaudio.com/DIY_HD_Opamp.htm)to replace the IC Opamps?

Click to expand...

Yes I have looked at the Burson opamps in the past and found them wanting measurement-wise, not to mention remarkably similar to AudioGD's offerings which are at a much lower price. Samuel Groner was my source, here : http://www.sg-acoustics.ch/analogue_audio/ic_opamps/pdf/opamp_distortion.pdf (its a big file). Perhaps they've got something newer now though as Samuel's paper was done in 2009 so I'll check out your link.

<edit> OK checked out your link - its chock full of dishonest marketing claims, so I'm not going to follow this up any further. A company that uses blatantly false claims like 'IC opamps are not designed with a focus on audio' isn't worth my time.

After all this tinkering inside, how does it sound?

In order to give this device a listen I decided on setting up a frugalphile system as my normal listening is done through highly customized active speakers which would be too impractical for anyone else to duplicate and hence render my listening experience less relevant to budding frugalphiles.

So in addition to the source, I assembled an integrated amp and speakers making the total system budget around $500, most of which is the amp which I bought at normal retail price.

The amp is a Xindak A06 which I must admit I succumbed to the temptation to modify. Not by much - just added additional power supply (LC) filtering and fixed up the lack of true star grounding. I haven't replaced any active components just added a new filter board and moved the main PSU caps off the amp PCB to reduce any possible interference from cap charging currents.



The speakers are the cheapest decent-looking passive ones from Taobao : http://item.taobao.com/item.htm?spm=a230r.1.10.1.3c8636&id=8581991365



Once again, I couldn't help myself and swapped out the ferrite cored chokes in the XOs - the tweeter one for an air-core and the woofer with an experimental inductor made from a sendust toroid. I'm not sure that these inductor changes made any audible improvements though, unlike the mods to the amp.

I was originally intending on posting a follow-up with my report from listening - but I got rather diverted from that mission, here's how.

Preliminary and quick listening comparisons showed not such a great difference between my reference DAC (a Muse 4 * TDA1543 modified to use 8 * TDA1387s) and the SD player's analogue outs as I'd heard through my actives so I was left scratching my head as to why the hot-rodded Muse had lost some of the life I was accustomed to hearing. Mulling over the internals of the Xindak my attention fell on the input opamp - a fairly lowly AD712. A quick check of its datasheet shows it has rather poor PSRR, meaning its sensitive to its power supplies to a large degree. Investigating how it was powered revealed the use of LM317/337 regulator ICs but rather poorly implemented. These chips do well enough in the audio frequency range but suffer from poor HF rejection so really need passive filtering on their inputs to give of their best - the designer hadn't realized this. Also they're rated for over 1A of output current - when feeding an opamp that requires just single digit mAs they barely hit their stride. Finally their physical distance from the AD712 was considerable, meaning long tracks between the two which tend to pick up noise.

If there's one general principle that I've found in designing good-sounding audio circuits, its this - keep loops as small as possible. If for some reason a small loop is infeasible, then make the largest part of that loop very high impedance. Power supplies normally need to be of the lowest impedance, so long loops of power supply wiring are to be avoided or minimized. However I wasn't up for physically moving the regulators or the opamp in this case, so I went for making the power supply wiring high impedance over most of its travel, by means of shunt regulators. The shunts themselves are a single device which looks just like a transistor and they offer a low impedance - less than 0.5ohms. They're also stupidly cheap, 2 cents or so and hence one of the frugal audiophile's best ever friends Because they're so compact they can be placed extremely close to the device they're regulating the power of. The outputs of the LM317/337 regulators I converted into high impedance - current sources - so that the long wires didn't matter nearly so much.

It turned out that while the LM317 (positive regulator) was happy with its new role as a high impedance current source, its complementary partner the LM337 (negative) wasn't. Fortunately current sources are androgynous, they don't care which polarity they work in so I junked the LM337 (they're more expensive too!) for a second LM317 wired backwards. The result was true sonic bliss - in one important aspect - soundstage depth - this set up now bested my actives.

Having discovered the audible delights of using shunts in the power supplies, I felt it would be unfair not to give the SD player's analog stage the benefit of these mods, so that now has received a pair of shunts and its series regulators are now current sources. A quick listen and I didn't notice much change but then its using LM6172s which are far less sensitive to power supply quality than the AD712.

As part of the power supply modifications, I fitted a socket to the AD712 so I could experiment with rolling opamps. This was rather instructive - the JFET types I tried had better soundstage depth and 'bloom' than the bipolar ones. But they also introduced some HF colourations to differing degrees. A case of picking my poison I guess...