Starlight Discrete DAC: a learning experience

You may find this post interesting or not. I just learned a lot with these tests and evolution of my design of the Starlight CD discrete DAC. Mostly, I want to thank Tom Browne for his patience and guidance throughout this interesting journey.

The Starlight discrete DAC has a relatively high output resistance: 10KΩ. With TTL-like levels it can deliver 1Vrms maximum. A commonly implementation of this DAC is with an output transformer in step-down mode (4:1) to reduce the output impedance of this DAC. The typical circuit is straight forward with a coupling capacitor between the DAC output stage and the transformer to block the DC current. The value of this capacitor is 2uF or higher. Many have used the Russian PIO with great results. All incarnations of this DAC sounded fantastic in my opinion, so there is plenty room for experimentation around this DAC.

I had a nice pair of LL7903 transformers. They are very nice and perform really well. These can be wired in 8:1 or 4:1. So they looked to be a right fit for this DAC. However, the high output impedance is a warning sign for any transformer as you would look to have a lot of inductance for a good FR. Higher inductance brings with it a higher parasitic capacitance on the winding which impacts the HF response. Achieving a high inductance and low capacitance is a challenge in any transformer design.

To confirm how good this transformer could be for this design, I measured the transformer to obtain its key parameters. To my surprise, it wasn’t that great. transformer model v2

Despite having a nice primary inductance (Lp), the capacitances are big. Looking at the FR response I found this not to be great (about 20KHz). Considering the Starlight CD player has LP digital filter, this will cause a slight loss of treble according to Tom.  At least 30KHz or more is needed to make this not an issue.

A workaround to this issue is the addition of a choke in the primary to compensate for the HF loss:

LL7903 response with choke


In the previous test, I used a CM choke to look at the impact of the added inductance. You can see that the addition of 65mH will extend the FR to 30kHz whilst the 130mH is way too much.

Playing with LTSpice, you can find the optimal value for this, about 50mH. Here is the simulation of the 65mH murata choke:

DAC with murata choke model SPICESo, how well this matches the real circuit?

DAC with murata choke model fit

Quite well. The transformer models are not easy to derive, but this one worked out well.

So problem is with a ferrite core is the distortion given saturation. Here is a view of THD for various choke options. The Murata clearly is not a good candidate as it’s adding significant HF distortion:

LL7903 THD response with choke


Geoff kindly sent me a pair of mu-metal C cores to wind the cores. I’m still waiting to receive new enamelled wire (i.e. 0.13mm) as needed 220 turns to achieve the inductance required and with the 0.3mm wire I have, it didn’t fit in the bobbin 🙁

After all these tests, I decided to get a pair of Sowter 9062, as they are the right match for this application. Starlight DAC with Sowter 9062

Given the longer wires of this transformer it proved to be more dificult to measure the parameters. However, we can clearly see why this transformer is a right choice here. The inductance is double of the LL7903, and the capacitance is low. The transformer should do at least 70-100kHz which is plenty for this use.

Given the long leads the FR above 80kHz-100Khz is a not right:

Sowter 9062 FR manual

Now, if we look at the circuit response at full tilt (4Vrms input to achieve 1Vrms output):

Starlight DAC Sowter 9062 THD and FRThe distortion is very low (lower than 0.01% for signal greater than 200Hz). The real distortion should be much lower as the input signal will be 4 times less. The FR is limited by my soundcard here. We can see that the distortion at LF is impacted by either the PIO or the transformer as it’s likely to saturate given the large voltage signal applied. If we compare the distortion with the LL7903, we can blame then the transformer and not the capacitor. I should re-run this test with a realistic input signal which is 1Vrms and not 4Vrms 🙂

I’m running out of time for these tests. I want to proceed with the build and am very keen to listen to one of the best CD players I’ve ever heard.

Some great experiments here that clearly show that what a good candidate seems to be good transformer for one particular design, it proved not to be the case after some measurements. Good to bear in mind 🙂

Stay tuned