Starlight CD transport and discrete DAC finished!

No Gravatar


After an incredibly long journey I’m glad to say that is nearly over. The Starlight CD transport (I2S output and 4x oversampling) with its discrete DAC sounds absolutely amazing. Getting back to play my CDs is what I was cherishing for a long time. I will be bringing my +500 CD collection from my previous home soon so that would be perfect timing.

I promise to write up a proper blog entry with a full description of the learning experience. For now, I will just sit down and relax. Listen to many CDs and enjoy!

All the acknowledgements and thanks go to Tom Browne for this fantastic design and all the help provided throughout this journey. I think this is the Starlight number 20 or so built so far. Not sure if there are other ones being built anyway.  Also my thanks go to Jon Finlayson for his help on the DAC boards, Richard for the UI stuff and to Tony for inspiring me to give it a go…




4P1L with screen as anode

No Gravatar

A friend in DIYAudio came up with a great idea of using the 4P1L in a different way for a pre-amp/line stage. Given availability of IT and its gapped current, he suggested connecting the 4P1L differently. Instead of using the anode as the anode element, the screen is used as anode. The 4P1L screen has a maximum current of 10mA and dissipation should be within the 1.5W.

The 4P1L curves in this mode

Let’s look at an initial transfer curve with Uak=100V:

4P1L Screen as Anode transfer









I wired the 4P1L in the following way to allow tracing with the uTracer:

4P1L screen as anode connectionThe connection is slightly different as the one suggested by Indra. If you are looking to implement filament bias, you will have to rearrange the anode and suppressor grid connection and expect a slight shift on the curves given the change in bias. Filaments are in series here, however is preferred to wire them in parallel when using this valve in filament bias as a smaller filament resistor will be required given there is twice the filament current when filaments are connected in parallel. This will help to keep the output resistance lower as the size of this resistor is smaller (remember it is reflected multiplied by μ+1 times.

Continue reading


Robustiano (V0.7)

No Gravatar

Hacked a simple PCB to build the follower to drive the 4P1L as suggested by Rod. I had to play with the LND150 setting resistor (R4) to achieve the 2mA of idle current. I ended up biasing the 4P1L rather hot at about 11.5W which exceeds the specs. The Q2 VBE was not possible to measure as the Q2 would oscillate I guess when I place the tester lead on Q1 collector and the voltage seems to drop when I try to measure it. Should have added a ferrite bead:

Robustiano v07 bench test

When measuring distortion against frequency, I was keen to see that the follower provided some impact in reducing the HF distortion. For example at 20kHz, THD reduced from 0.96% to 0.59% @1W output power and from 7.84% to  3.52%, that is close to half the distortion I had before:

Robustiano v07 THD tests

What is nice to see now is the effect of the follower providing sufficient source current to the 4P1L grid. Above 2.5W, the grid current kicks in and we can see how “Robustiano” can deliver 3W at less than 1% until starts to clip about 3.5W:

Robustiano v07 THD versus power

I found that if I were to reduce the Rf further and therefore increasing the collector current but obviously exceeding the 4P1L power dissipation too much as collector current was about 45-48mA, the distortion at 20kHz falls significantly. I suspect I should increase the collector current to enable better drive of Q2 due to its Cib (30pF). To keep the current feedback arrangement this could be done by reducing the negative emitter voltage source (V1). Should try this I guess…






Robustiano (version 0.6)

No Gravatar

It was now the turn to the BJT to show what it can do in this circuit. Here is my quick breadboard with components I had at hand, so there is a slight variation from the simulation:

Robustiano 4P1L v06 BJT testThe Q1 is obviously an MPSA42. C3 was substituted with a 10uF/25V electrolytic (yes you read well). Rf is a 250k carbon pot and R2 is built of a series of 1K pair in parallel plus 100 ohms in parallel, four resistors are wire-wound 1W. C1 is actually a 30uF/450V ASC Oil cap.

I haven’t measured the collector current but it seems to be around 1.3-18mA. Variance is due to tolerance of RE and R3. I need to measure the actual resistance of Rf but is somewhere between 180 and 200K.

As Rod suggested, the circuit is very stable and easy to dial the right feedback with the potentiometer. The 4P1L is biased about -9V and is slightly under the max Pa in this case.

Distortion is as predicted in the simulation (e.g. 0.1% for 1W and about 0.2 – 0.25% for 2W). Just above 2.2W distortion creeps up rapidly given grid current, which is not modelled properly by the 4P1L pentode model:

Robustiano v06 THD versus powerLooking at distortion versus frequency, it’s interesting to compare the BJT performance against the depletion FET. With lower Cob compared to the Coss of the FET, the BJT should be able to drive better the 4P1L. In fact, the BJT is more linear when swinging many volts compared to the depletion FET, so the proof is in the pudding:

Robustiano v06 tests THD The BJT is indeed more linear but if we compare the THD vs frequency of both drivers when providing 2W output power, we can see that the BJT is suffering as much (and even more given poorer slew rate) than the FET at frequencies above 12kHz. Also FET’s THD versus frequency is more linear up to 10kHz, whereas the BJT has a peak around 6kHz and a dip closer to 10-11kHz. Either way, the BJT outperforms the FET in overall THD up to 11kHz.

I need to listen to this circuit now…