Transforming the Stax SRD-7

(updated 12/26/22)

A few months before Covid-19 arrived in the US, I became interested again in listening to music via headphones, an interest which I had mothballed a couple decades prior. As often happens, interest attracts opportunities, and shortly thereafter I was fortunate to acquire a set of fully-functional Stax SR-X MkIII headphones with SRD-7 adaptor from their original owner. My prior experience with electrostats was the Fontek A-4 electrets (which I still have), and I always liked their clean, unboxy if bass-shy sound.

Based on what I’d read online, expectations were high when I first listened to the SR-X/SRD-7 combo. And boy was it disappointing. They sucked! Aggressively bright and forward, with no warmth or spacial sense at all. How could such a product have gained a stellar reputation as a “monitor” or “reference” headphone, and drawn high praise from so many golden-eared audio reviewers?

After confirming that the SR-X were indeed working properly, I decided to look into the SRD-7 for obvious faults and possible improvements. There isn’t much on the interwebs about improving this energizer. A guy in Iceland used to repackage the SRD-7 transformers into his own enclosure with different bias networks, sold it for beaucoup bucks, but gave no measurements or data to back up or show his work. Others suggest the usual “replace the wiring and capacitors”, and bypassing the selector switch. The latter is a good idea, but not transformational.  So lets take a closer look at it. Here’s the schematic.

The signal first encounters a component labeled “PTH” and identified as a Murata BD4R7M.

Looking like a large, old-school disc capacitor, it is actually a thermistor of nominal 4.7Ω, a “safety” device whose impedance increases in proportion with the current through it. This is nuts – a level-dependent impedance modulator on the input! This can’t be good…

And it isn’t. In-circuit, the distortion through the PTH with only 1V RMS appplied is over 0.5% ! Really awful. Definitely get rid of it.

Here is the SRD-7’s input impedance after replacing the PTH with a 4.7Ω 5 Watt wirewound nonmagnetic resistor. The channels are well-matched. This is a very easy load to drive, more than 20Ω over most of the audio band. A few good watts will do just fine. This is definitely a case where amp quality is more important than quantity (of power).

The following graphs are detailed (0.5 dB per division) looks at the SRD-7 frequency and phase response with SR-X headphones attached, referenced to the amplifier output. The yellow curve below is the stock setup with 4.7Ω resistor replacing the PTH. As you can see, it is not very flat. Let’s see if it can be improved.

The 4.7Ω resistor is followed by 27Ω across the transformer input. I’m not sure why Stax chose this configuration, it needlessly wastes some of the input voltage. It makes more sense to put  the terminator before the series R. The blue curve above shows the result of doing just that. About 1.5dB of level is recovered, and the low-to-high response variation is reduced by half. A win-win.

Back in the 1980’s I learned about transformer termination and compensation from Deane Jensen of Jensen Transformers (may he RIP). FWIW, I can tell you, Deane would not be very impressed with the linearity of the transformers in the SRD-7. But all is not lost… If we can assume that a low-output-impedance amplifier (i.e. solid-state with feedback) will drive the SRD-7, then the source and load impedances are well-enough defined that we can optimize compensation networks for the transformer input and/or output to flatten the response.

After some experimentation it became clear that it would be best if this were a two-stage affair; the first stage levels out the overall upward frequency response slope, and the second (optionally) tames the high-freq peak.

For the first stage, I eventually arrived at an impedance-compensating network of 619kΩ + 1000pF across the xfmr output. This network not only levels the response curve, it also lowers harmonic distortion through the transformer across the spectrum. The yellow curve below shows the result.

The second stage filtering is a gentle (less than 1dB) high-frequency shelving above 2kHz and can be accomplished in several ways; with basic tone controls if your system has them, with digital EQ, or with a passive filter between the amp and the SRD-7. Here’s how to do it passively.

Looking at the freq response and impedance curves, notice that they are a mirror image of each other above say 1kHz. So adding a parallel RL network in series with the input will flatten the response peak. It doesn’t take much; a 56uH inductor in parallel with 1Ω is just about right. The blue curve above shows the combined result of the input + output compensation. This gives the most neutral frequency response, suitable for use with the widest range of Stax ‘phones.

For the naturally-bright SR-X MkIII, you’ll want to correct their tonal imbalance as well. Again, digital EQ or preamp tone controls will do the job. Boosting the bass tone control by just a couple dB and lowering the treble to taste makes for a very enjoyable listening experience.

Because the SR-X brightness is a broak peak centered in the upper-mid treble at around 4kHz, a more accurate approach is to use digital EQ, such as the excellent (and free) Equalizer APO ( Equalizer APO ). After installing, rename its original Config.txt file to something else and create a new one in the same folder with these two lines:

Preamp: 0 dB
Include: srx.txt

Now create and save another text file in the same folder named Srx.txt with the filter details:

Filter Settings file
Notes: ~ 4kHz Notch w/ gentle bass boost for modded SRD-7 w/ SR-X MkIII
Equaliser: SR-X MkIII
No measurement
Filter 1: ON HS 6DB Fc 185.0 Hz Gain -2.5 dB
Filter 2: ON PK Fc 4000.0 Hz Gain -3.5 dB Q 0.55
Filter 3: ON HS 6DB Fc 8000.0 Hz Gain -0.8 dB

With these settings (without the passive filter), the SR-X Mk3 and modded SRD-7 are a remarkably good system. If the passive filter is in the system, then set Filter 3 to OFF.

Off-topic but worth pointing out: Every amp and xfmr coupler for electrostatic headphones I’ve seen has a single source of bias voltage fed to both headphones. A consequence of doing this is crosstalk through the bias network. The SRD-7 definitely has it (see graph below); -44dBr at 10kHz, and rising. Also notice the bump around 700Hz, which corresponds to the zero-phase crossing point in the impedance and freq response curves, where induction through the xfmr is most efficient. This sort of crosstalk/leakage creates a level- and frequency-dependent spatial smear, tending to pull strong high-frequency content from either channel toward the other. It’s not a deal-breaker; every amp/coupler has it to some degree, and some designers specifically add crossstalk to change the perceived soundstage . But it is definitely present and noteworthy.

And that’s it. Here is the new circuit sans bias stuff. Quite simple, really. Even simpler if you don’t use the L1 || R2 passive filter.

And here is the total harmonic distortion of the modded SRD-7 at 1kHz with 100V RMS output, 0.0077%. That’s lower than the SRM-007’s 0.01% rating. This output level corresponds to 99dB SPL with the SR-X cans. Comparing this plot to the earlier one of the PTH gives you a good visual indication of the magnitude of improvement these mods deliver.

 

Parts quality, sourcing, and installation

I highly prefer and recomend using nonmagnetic (NM) resistors and capacitors, especially when they’re located in the pulsating magnetic fields of transformers. But they’re becoming more difficult to find.

Coherent stereo soundscapes depend in part on both channels being identical. High-stepup-ratio transformers like those in the SRD-7  are very sensitive to the impedances at their inputs, so it is important to match the components for each channel. One way to do this is to buy twice as many as needed and match them yourself.

For decades I’ve been recovering quality components from electronic equipment being scrapped/recycled. For this little project I was fortunate to have everything needed on hand. Specific notes/suggestions on sourcing the parts follow. As of this writing (August 2022) Digikey has them in stock (link to the BOM: https://www.linearz.com/wp-content/uploads/2022/12/SRD7BOM.xls).

After removing the PTH’s from the pcb you’ll see that mounting holes for the 4.7Ω resistors are already present, and are even labeled on the foil side! This tells me the Stax engineers knew that the PTH’s were cr@p.

All of the cement aka “bathtub” wirewound resistors I’ve encountered from the 70’s and 80’s (including the ones in the SRD-7) have been nonmagnetic construction. Nowadays most wirewounds are not NM and the materials used are rarely specified on the data sheets. All of the parts on the pcb are sitting right next to the transformer windings and would really benefit from being NM. The 4.7Ω bathtubs I used are from the 80’s and are NM.

R3 : 4.7Ω 5 Watt wirewound resistor. The Bourns FW50A series may be the best choice; it specifies tinned copper leads but doesn’t mention the end cap material. Part number FW50A4R70JA.

Relocating the 27Ω resistor to the input side doesn’t require any mods to the pcb. Simply swap the input and output wires at the pcb for each channel. (A bit OT, but while you’re at it, replace the 1uF 160V electrolytics capacitors, preferably with NM film types. I used ERO 1uF 250V mylar films which are NM. I also replaced a couple of the carbon film resistors that had drifted high in value.)

The output RC network is best installed right at the output jacks; there isn’t room for them on the pcb. In use there could easily be 200 VRMS AC across these componenets, so they must be rated accordingly. I used a Wima polypropylene capacitor and Vishay/Dale RN60 resistors.

C1 : 1000pF 400VAC Polypropylene capacitor. The Wima FKP-1 series is NM, a good choice and is widely available. Part # FKP1O111004B00JSSD

R5 : 619k (620k) 1/2 Watt or better metal film resistor. I used Vishay/Dale RN60 237k + 383k in series but a single 619k part will be fine. The TE RR series has copper leads and will work fine. Part # RR01J620KTB .

If you’re not using the L1 || R2 second-stage filter, you’re done and can skip to Summing Up. For the 56uH inductor, I had on hand some 58uH coils wound on a ferrite rod core and used those. Anything in the 55-60uH range with low series R is fine; just make sure the pair you use is well-matched.

The L1 || R2 combo doesn’t have to be installed inside the SRD-7 case. But if you go that route, it’s probably best to install them in the rear of the enclosure, spaced away from the case and orthogonal to the transformers so as to minimize the magnetic interaction with them. I chose to install mine inside my power amp and gave them a separate set of output terminals. This gives me the flexibility of using the SRD-7 with various Stax ‘phones, with or without the input RL.

L1 : A cored inductor is preferred due to their lower series resistance. Use low-permeability ferrite or Sendust cores; no iron or iron dust for audio. The Bourns 5900-560-RC looks like a good choice, 56uH with ferrite core and low DCR.

R2 : 1Ω 2-3 Watt wirewound resistor. Most modern wirewounds have magnetic leads and endcaps; even the Vishay RS2 I used do. The Stackpole WW3FT1R00 datasheet specifies copper leads so we’ll go with that.

Summing up

I went into this project with a strong prejudice against using transformer coupled outputs on audio power amps. In this case, they not only work well, they can outperform direct-coupled HV amps in many respects. These are not just subtle improvements. With these changes and a good amp, the SRD-7 now sounds MUCH more natural and transparent, especially with the SR-X MkIII ‘phones using the digital EQ. The important midrange is superb, my hands-down favorite of all headphones. The abrupt rolloff at 20kHz is still there (it always has been, but was masked by the high distortion) and  does take a little “air” out of the soundstage, but like the crosstalk, it is not a deal-breaker. The “transformed” SRD-7 is very enjoyable and comfortable to listen through, and these mods are very worthwhile, easy to do, and will outlast you. And if you change the bias circuit to provide both high- and normal-bias voltages, you will have a versatile, great-sounding adaptor that works with all Stax headphones, and for very reasonable cost.

Enjoy,

John Bau

Questions? Comments?  Contact