Modding the DirectStream DAC MKI

Thanks Ted! I meanwhile noticed that I already had R5 and R6 changed…my assumption that it was only R6 was wrong.

So now I just need to have the Tantal soldered over R6 finally. Maybe I wait until JK tells if C704 has a critical ringing effect measured with the LL1948 to decide if it stays.

1 Like

В моем случае LL1948 с C704 звучат лучше. Без них звук на средних/высоких частотах был суховатым и резким. Но было бы интересно увидеть такие же измерения с LL1948. Спасибо jkrichards.

1 Like

I think it would be good to know. If we get time we will run the 1948 through the AP.


We will have measurements of the Edcor 150 150 (stock), XS4400, and LL1948 with and without C704 installed this weekend.


I have the Vishal caps now installed over R6. Not burned in, bass is immediately cleaner and deeper, the rest still needs to open up a bit, currently the top end is a little more recessed than without them, hope it comes back after burning in this soldering and cap again, but the first effect otherwise is promising.

Transformer sweeps as promised.
See notes at the bottom of each sweep. With and without C704.

Edcor 150-150 (Stock)

Edcor XS4400
Both 1:1 and 1:2 modes were tested (see plot titles). Note the fairly significant phase shift and attenuation at 20k (-.6dB) (even without the 15nF capacitor) when wired in 1:2 mode. (MKII)

Here are the sweeps for the LL1948 transformer. The configuration is described in the notes field for each. These are the best of the three however the nickel core transformers will out perform these by a large margin.
All measurements were done with the transformer wired 1:2 (parallel primaries, series secondaries with grounded center tap). 1st one is without C704. 2nd one is with C704. Notice the high frequency, distortion, and phase change with C704 installed.

Our transformer design goal is to improve the high-frequency errors present when using both the XS4400 (mk2 DAC) and LL1948 in 1:2 mode while maintaining a low THD at low frequencies by use of a high-quality nickel core.


Thanks! But I can’t read the diagrams…does it mean any danger of ringing with C704 and the LL1948?

1 Like

No ringing with the C704 using the LL1948 however the High Frequency is down almost -1.5dB at 20k. Without C704 there is HF ringing. I would say Keep the C704 in for the LL1948 even though you loose some HF (blue) and the distortion is higher at HF (red). The worst measurement is the phase error above 1k with the C704 installed. Last 2 photos under LL1948.

These issues will be addressed and corrected with our new nickel core transformers. We will recommend NOT using C704 with them.

@tedsmith Are the values correct in the opamp schematic that was posted here? We noticed a fairly aggressive 2nd order LP filter was used with the values listed. This also caused some HF phase error which I understand is normal when using these types of filters.

1 Like

So another value of C704 couldn’t change that?

Didn’t acoustically notice HF loss with LL1948 and C704, but noticed it with the cap over R6 (at least when not burned in).

You shouldn’t have HF loss with the VOCM mod however the bass should become tighter and more correct.
Are you sure you installed them caps with the correct polarity?

Should be yes, I’ll verify…maybe they have to burn in. Bass is better immediately.

Also if C704 is installed keep your XLR cables as short as possible. > 1 M

Maybe a 12nF cap would be a better value.

I should have been more precise regarding top end:

The thing I could compare was the XS4400 without C704 with the LL1948 with C704, R5/R6 replaced. Here the latter combination fortunately had a more open and pronounced top end, inspite of the -1,5dB C704 effect you mentioned. Probably because the LL1948 is so much better than the XS4400.

Then I had the VOCM cap built in over R6 and so far, besides bass much better, top end got a little more recessed (but I wait with final judgement as I always have big differences of burn in).

If you now say 12nF could be better, I think I wouldn’t have all the soldering done again without being sure it really measures differently regarding top end. There’s still chance to react with cabling anyway.

But every tweak to get some better top end openness (except switching transformers once more) is welcome generally.

The caps are built in correctly as far as I saw (+ on the right).

Are you sure? The cap over R6 should have around 47µF (=47000nF) and the cap for C704 should have around 0,015 µF (=15nF), right?

Sounds promising! Will you be installing your transformers, or just sending them out to be installed?

You’re correct, I confused values and caps in the text here (but built in correctly :wink: )
I now also got that JK means the C704 with the 12 vs 15 nF discussion.

1 Like

I now listened again more in detail and discovered that I had kept a wrong speaker setting from before my holiday.

So now bass improvement is even more obvious and the top end discussion very minor, I guess it will burn in.

The bass improvement with the VOCM mod alone is absolutely essential for me, like a new DAC there. By the way it also gets little smoother in the mids, like certain tiny resonances avoided, stage bigger in all directions. But the bass, oh man, it’s crazy. There might be some who were used to the previous less controlled bass who might miss a bit, but for those with a well balanced setup, it’s a revelation…can’t wait until it’s burned in and I have my other treatments done again…wow.

But I also saw I have to take care if the DS sounds different open and closed.

W/O C704 and R6 I could not listen to the LL1948. With and a bit of breakin (yes it took like 100 hours) all is wonderful. Mr Richards knows what he is doing.


Another issue we noticed is the fact that the cobalt core LL1948 has higher distortion than the steel core XS4400. Maybe because of the small core size of the LL1948 :thinking:
This is the reason we are using a large nickel core in our transformer design. It’s more expensive however we feel it’s needed in this application. It improves the high frequency issues and lowers distortion by a large amount.