Upgrading the Stax SRM-252S
– aka –
Coaxing a Quart from a Pint-Sized Pot
Since Covid’s appearance, I’ve been mostly in headphone listening mode, spending more time at the computer investigating music and less time doing more physical and social activities. This was furthered by the rediscovery of a set of Fontek electrostatic headphones which I had bought in the mid-80’s and stashed away in a closet shortly after Y2K. Hearing the Fonteks whet my appetite for ‘stats again. Shortly thereafter, life presented me with a set of Stax SR-X Mk III with SRD-7 adaptor from the original owner. I was not impressed. (Read more on this here .)
So I decided to try one of their current offerings and got an SRS-3100 system, which consists of L300 headphones with an SRM-252S amplifier aka driver unit.
I was pretty underwhelmed with the sound of the 3100 system. Sure, they have the “open” quality characteristic of planar ‘phones, and are fairly smooth at high frequencies. But overall, decidedly sterile and unengaging. Certainly not worth nearly a grand!
Looking at opinions/reviews online, I found widespread dissatisfaction with the 3100 system, many opining that the L300’s were ok and the 252S amp was the main culprit. I decided to investigate further. I found some prior attempts to improve the 252S, mostly just “upgrading” the capacitors. The most thorough was by a poster named “borizm” on Head-fi.org, who did a quite good analysis of the very-similar SRM-252A circuit. Most of the 252A-to-252S changes were replacing discontinued transistors, some layout changes, and a different parts numbering scheme. (That borizm did his analysis without the aid of measurement gear is a credit to his discrete circuit chops.) He correctly concluded that the central issue with this amp is the HV power supply; it has limited available current and high, nonlinear output impedance. This is a direct consequence of the tiny box; there is simply not enough room to put a really good +-210VDC supply inside.
I find that the 252S has four main problem areas: The 12V DC power source, the HV power supply, gain structure/feedback, and parts quality.
The 12VDC wallwart
The wallwart that comes with the unit is a poor choice for a DC power source. It is unregulated and has over 70mV RMS of ripple, which is presented to the SG3524 PWM IC and passed through unregulated to the entire circuit. This is an easy fix. Replace the stock wallwart with any decent regulated power module. One with low noise and low, linear output impedance would be ideal, but even one of the ubiquitous (and cheap) 1-2 amp plugin switchers will clean this up nicely. An example of a very good one is the HG Power ADPV26B, 12V at 2 Amps. This unit has low noise across the audio band (ADPV26B noise spectrum), better than many desktop switchers. (HG Power is an OEM, the ADPV26B is sold under other brand names, such as Insignia, so look around.) Whatever you use, you will obsolutely need to swap the wires to the plug so the +V is on the barrel and -V is on the center pin.
HV Power Supply
As stated earlier, it would be nice if we had more current to work with. But attempting to increase the existing circuit’s power output would also generate more heat internally. With any Class A circuit, heat disssipation is a major concern. In stock form, the 252S enclosure feels warm but not hot to touch. That is a good thing. So we’ll leave it as it is.
The output impedance of the HVPS is determined by the filter circuit following the pulse transformer. The stock circuit uses two stages of 470Ω with 22uF to ground.
These do a decent job of smoothing out the approx. 65kHz pulse trains into DC. But the output impedance is very high at audio frequencies; 7.5Ω at 1KHz, almost 80Ω at 100Hz, and climbing rapidly below that. It doesn’t get below 1Ω until above 10kHz! It is plotted in the yellow traces below.
An obvious improvement would be to use larger values of capacitance, say 1000uF or more. But 1000uF 250V ‘lytics are quite large, and there is simply no room for them inside the teeny enclosure.
Borizm’s idea was to replace the second 470Ω with two 470uH inductors in series, and replace the capacitors with 47uF. He likely chose 1mH total inductance because it’s impedance is about 470Ω at 65kHz. The curves for his configuration are plotted above in the cyan traces. The impedance is lower, but the phase is even more nonlinear, with obvious resonances right in the heart of the midrange. I tried this circuit and, while it is better than stock, the midrange tonality had become “boxy” or hollow, very unflattering to vocals in particular. The cause is pretty obvious: Adding the inductors created a resonant RLC circuit. One can’t just arbitrarily pick L or C values; they interact and must be carefully chosen to minimize resonances in the audio band. This resonance is not visible in the impedance curves, but the impedance phase reveals it quite clearly.
I came up with two optimized RC+LRC filter circuits that satisfy the criteria of lowering the output impedance with a smooth, resonant-free Zphase response. The first circuit I call the “tidy” version, optimized using the largest capacitors that will fit comfortably on the pcb. It is shown in the yellow curves below. The second circuit, the “optimum” version, is optimized for the largest capacitors that can be safely crammed inside the case. It has the lowest Z and Zphase transition that can be acheived in the available space with currently-available parts (cyan curves). And it is by far the best-sounding one.
The only downside of this version is, two electrolytics are mounted unconventionally on top of other components, and must be spot-glued in place to be secure. Easily done.
In the “optimum” schematic below, the inductor and resistor that replace R73 & R74 are shown as separate components. In reality, some of that 2.4Ω is in the inductor winding, and the resistor after it makes up the difference. So the actual resistor used is 2.4Ω minus the inductor’s DCR. The actual resistor value for the recommended inductor is given in the parts list.
Gain Structure / Feedback
As delivered, the 252S has oodles of gain; 770X or +57.7dB . According to the manual, it is designed to give 100V output with inputs as low as 125mV. Why did Stax do this? I don’t know of any audio source with an output level that low. It is even lower than the old consumer electronics standard of -10dBV (316mV). And it is 16 times lower than the 2V RMS output of most modern DACs. That’s a lot of signal level to throw away right at the input level control, which for me is typically set at around 11:00 for comfortably-loud listening. Why not open the control a bit and operate the amp at lower gain, increasing the feedback and lowering distortion?
The gain is set by the feedback resistors 300k / 390 Ω. I first tried increasing the 390Ω shunt R to 620Ω (the value used in the 252A). The sound became colder and harder. Apparently the FETs prefer the higher bias at 390Ω. Lowering the 300kΩ feedback R will require more current from the output. So I incorporated one of borizm’s suggested changes, but for a different purpose. Changing D7 & D8 from a red to a green LED gives the output stage about 200uA more current. He wanted it available for output drive; I’ll use it to drive a lower feedback R. I first tried 237k, then 200k, and finally used 150kΩ feedback R’s, which halves the gain (from 770X to 385X) while using about 220uA more current at max output levels. The result was a huge cleanup of the sound across the board. Along with the HVPS mods, the 252S was now sounding more transparent, and the soundstage more coherent.
Parts quality
Slide the cover off and look inside. The circuit is nicely laid out and quite efficient i.e. dense. There is very little room to spare. The film capacitors Stax used are good enough for their purpose. But see all the tiny resistors? I did not expect to see 1/10 or 1/20 watt resistors in a high-voltage amp. And they all have steel endcaps and leads.
My main concern is that some of them are operating at or above their maximum rated voltage. And such small resistors have poor Vcr (the change of resistance with applied voltage). We’ll replace all of the ones in the signal path with 1/8 or 1/4 watt units with a higher operating voltage and lower Vcr. I used mostly Vishay/Dale CMF55 or RN55.
The feedback and output resistors are the most important quality-wise. They should be 1/4 watt or better. I used Dale RN60 (300V rated) and RN65 (350V) for those. I was shocked to see that Stax used the little 1/10W resistors on the outputs. What were they thinking? You’ll see in the parts list that I specify 4.99KΩ for the outputs instead of the standard 5.1kΩ. I tried both and preferred the slightly warmer sound the 4.99K gave.
Parts removal and replacement
These mods apply to the pcb labeled #PB-201. As do the part numbers in the parts list below.
The board was assembled at the factory using lead-free solder, which has a higher melting temperature and doesn’t “flow” very well, making desoldering more difficult. A vacuum desoldering gun helps a lot, but even so, several times I ended up adding some leaded solder to a joint to aid its removal.
After removal, it’s best to install the small parts first and large capacitors last.
A few parts, specifically C39 and the series LR that replaces R73 and R74, must be mounted on the back side of the pcb, to make room for the larger capacitors up top. Before committing to them, make sure these parts will not contact the bottom of the case when it is slid into place. Taping an insulating sheet to the bottom is not a bad idea.
Here’s what the finished board looks like with Optimum mods installed. As you can see, the large ‘lytics just fit, with little rooom to spare.
After you’ve installed the new parts, let the unit warm up for 30 minutes and adjust the DC balance on each channel (TVR3 and TVR4) for zero across the diff output resistors, and DC zero (TVR1 and TVR2) from either one of them to ground.
Summing up
These changes, and a few minor ones noted in the parts list below, address all of my complaints about the sound of this amp. These mods deliver a HUGE improvement over the stock amp, giving most users what they hoped they were getting when they first bought the system. I can now comfortably say that the SRS-3100 system with these mods is among the best headphone systems I have heard. The main problem I have with it is, the sound is now so clean and smooth that it is VERY easy to listen to things way too loud.
Is there room for improvement? Probably. There is still plenty of excess gain in this circuit. IF a little more current can be given to the output transistors, the feedback R could be lowered even more, thereby lowering gain and distortion further. And IF larger-value 250V capacitors ever become available in the case size needed, the HVPS filter circuit could be re-optimized, lowering the output impedance even more, and improving low bass integrity a bit. Those are both big IF’s; for now, this is as good as it gets, and it is very good. A quart from a pint-sized pot.
Parts sourcing
I highly prefer using resistors and capacitors that are nonmmagnetic (NM). But they’re expensive and getting harder to source in thru-hole. Fortunately, Digikey has the HV capacitors and critically-important feedback and output RN60’s in stock. Here’s a link to the BOM for the Optimum version. Digikey has everything in stock as of this writing (Sept 2022). Note: A few Yageo resistors are specified in the BOM. They are NOT nonmagenetic; but they are 1/4W parts rated at 250V and have copper leads.
Parts list and details
“NM” means Non-Magnetic
“=” means same value
Part # Descrip Original New Details
------ ------------ ---------- --------- ------------------------
R1,2 MF resistor 68K = NM
R3,4 MF resistor 3K 1K Dale RN55D1001FRE6
R5,6 MF resistor 33K = Dale RN55D3302FB14
R17,18 MF resistor 390R = Dale RN60D3900FB14
R19,20 MF resistor 390R = Dale RN60D3900FB14
R23,24 MF resistor 100R = NM
R27,28 MF resistor 300K 150K Dale RN60D1503FRE6
R29,30 MF resistor 300K 150K Dale RN60D1503FRE6
R31,32 MF resistor 1K = NM
R33,34 MF resistor 1K = NM
R39,40 MF resistor 2K4 = NM
R41,42 MF resistor 2K4 = NM
R43,44 MF resistor 750R = NM
R45,46 MF resistor 750R = NM
R49,50 MF resistor 5K1 4K99 Dale RN60C4991FRE6
R51,52 MF resistor 5K1 4K99 Dale RN60C4991FRE6
D7,8 LED Red Green
C1,2 Cer cap 330pF 100pF NP0/C0G
C31 EL cap 1uF 2200uF Rubycon 16ZLS2200MEFC10X25
"Tidy HVPS filter"
R71,72 MF resistor 470R = NM
R73,74 MF resistor 470R 120uH L with
3R0 series R 250V rated
C7,8 Film cap 10nF 120uF Nichicon UCY2E121MHD1TN
C37,38 EL Cap 22uF 56uF UCC EKXJ251ELL560MK20S
C39,40 EL Cap 22uF 10nF 250V film
"Optimum HVPS filter"
R71,72 MF resistor 470R = NM
R73,74 MF resistor 470R 150uH Abracon AIAP-01-151K-T with ***
1R0 series R 250V rated
C7,8 Film cap 10nF 180uF Nichicon UCY2E181MHD
C37,38 EL Cap 22uF 100uF Nichicon UVK2E101MHD
C39,40 EL Cap 22uF 10nF 250V film
*** Abracon specs the 150uH inductor as having 1.2Ω series R. The ones I got both measured just under 1.4Ω, hence the 1Ω buildout R for 2.4Ω total.
Enjoy,
John Bau
Questions? Comments? Contact