Why Should I use a Coleman Regulator to heat my directly heated triodes? (Rod Coleman)
I asked Rod for some details around DC in DHT and kindly offered writing this page. Here is Rod’s explanation as to why is better to implement DC regulation in DHTs and also what type of regulators are better suited for audio…
Why don’t I just use parts in my junk box?
To answer, let’s look at:
Alternative Heating Methods – A brief roundup of approaches to Filament Heating, with attention to their weaknesses:
AC heating does have one advantage compared to any other – it is very low cost to build.
However, it is a false economy in every case, because the sound is far worse than properly implemented dc, and you can risk higher running costs, because of loose tolerances on mains supplies these days. For instance, here in the UK, the tolerance has recently been loosened again – to 230V +/- 10%. Yes 207V to 253V, incredible but true. Running DHT filaments at +10% will reduce the lifetime of DHT substantially.
The 4 big sonic problems with AC:
- Substantial Intermodulation Distortion (IMD). The large ac voltage swing across the filament causes corresponding changes in the grid-cathode voltage. The (constantly varying) filament voltage also changes the effective localised anode-to-cathode voltage, both effects producing hum and “second-harmonic of hum” cross-products in your music signal. This just isn’t hi-fi.
This calls for a humbucker pot to be used to null 50/60Hz currents of opposite phase. This does not completely work, because harmonics of the mains waveform are generated in the DHT….. Interestingly, this harmonic phenomenon has been examined by Dmitry Nizhegorodov, whose work is well worth a read “On Correlation Between Residual DHT Filament Hum and AC Frequency. Distortion-induced hum in directly-heated triodes”.
The size of the problem is directly related to the AC voltage. The problem is quite noticeable when the AC voltage is 5V, and leads to humm that can’t be entirely nulled out.
- With AC, the only mains noise-blocking method you can adopt is chokes – which have only a minor and band-passed effect. Mains noise, including coupled noise from B+ rectifier pulses, mostly marches right in.
- The filament voltage gradient (constantly varying if AC-heated) produces a music-signal voltage across the filament. If ac-heated, this voltage drives a music-signal current through the transformer’s secondary, where it sees the parasitic inductance of the winding, and the stray-capacitance to ground. An unwanted EQ-network, in other words, or perhaps a tuned-RF sniffer coil, according to its properties!
- The cathode current (ie music signal) is flowing in the same internal wiring (of the DHT) and through the same pins as the filament current. Therefore, if the filament supply carries current noise (as opposed to voltage noise), this will be DIRECTLY mixed with your music signal. If your 1A supply has 0.1% current noise, that’s 1mA. But think about what 1mA of noise does to a 60mA 300B cathode current, or 1mA in the 6mA cathode current in a 26 DHT…
My Regulator kits are designed to tackle all of those four problems. The design has a special focus on blocking HF noise. Judging from what folks around the world are saying about the sound, it seems to work.
DC heating with LT1084 or LM317 in current-source connexion
These are cheap IC voltage regulators with noisy 1.2V bandgap references. They are designed for supplying voltage rails in consumer and industrial products. By connecting them so as to force the reference voltage across a fixed resistor, they can operate as a current source.
But in practice, these cheap regulators do not work well for filament heating, for a number of reasons:
The bandgap reference used in almost all IC voltage regulators is the bandgap type. This is cheap to make, but notably noiser than high quality buried-zener references.
Noise is a vital consideration for filament heating, since these regulators (when connected as a current source) impose the noise on the filament current. But the filament current runs in the same metal wire as the cathode current in a DHT! Therefore, regulator noise is directly added to the music signal, causing degradation!
For LT108x, (LM317 is similar), the data sheet specifies the noise as:
0.003% of Vout, Typical, 25 deg C, RMS value, 10Hz-10kHz.
When connected as a CCS, for say a 1,2A 300B filament, the regulator applies 1,2V across a 1-ohm resistor (the 1 Ohm is the current-sense resistor), so the output noise is:
Vn = 1,2V x 0.003% = 36uV;
and the noise current is translated by 1-ohm directly into amperes: 36uA.
So the CCS injects RMS noise of 36uA into your filament, where it is directly coupled to the cathode current.
Compared to full-scale dc current is class-A amplifiers:
- For 300B at 60mA, this is 0.06% rms
- For 26 preamp at 6mA this is nearer to 0.5% (1.05A filament).
Now consider the noise as a percentage of the actual music signal: 0,3% of noise for a 10mA signal in a 300B, or 1% of noise for 3,6mA of signal current in a 26 preamp!
No wonder the difference is audible! And the static noise of LT1084 has been observed and reported in the “26 preamp” thread, by DHT Rob.
Remember, these are RMS values, so the peaks are bigger. For some samples of the chip, the noise will be worse, as it is not guaranteed.
And, if the test bandwidth were properly opened up to audio amplifier limits, the noise would be substantially worse.
2. Suitability of LM317 internal circuit. (LT1084 performance is comparable).
The LM317 datasheet shows the internal schematic. Look at the large number of parallel paths from input to output! Of course these are entirely necessary for the IC’s usual application: voltage regulator, with good dc regulation. But such a complex circuit is not required for programming a current – and the complexity has many disadvantages.
Now look closely at the “typical performance characteristics” curves. “Ripple rejection” is a the ability to maintain a stable output, even when the input voltage is subject to noise and mains-related ripple. At first blush, it looks good – over 80dB at 100Hz! Now look at the ripple-rejection versus current. At 1,2A for a 300B, the rejection has already dropped to about 60dB – not so good. Then, it gets worse still. Consider the broadband noise found on mains supplies everywhere now, and more importantly, the current pulses caused by reverse recovery of rectifiers combined with stray inductance in the rectifier wiring. All of this noise assaults the filament regulator with bandwidth extending into hundereds of MHz. How does the LM317 fare with this? Look at the curve of Ripple Rejection versus frequency. The LM317 is consistent up to only about 10kHz, where the rejection falls sharply. At over 250kHz (for 1,2A) the ripple rejection has mostly vanished.
Conversely, the Coleman Regulator, with passive first stages of regulation and short-loop current programming, holds its rejection for decades of frequency further.
OK, what about the acid test – listening? There have been a number of independent comparisons made by builders upgrading from LT1084 based solutions to the Coleman Regulator, so you don’t need to take my word for it. Some examples:
It is difficult to see any advantage to be gained from RF filament heating.
In the first place, the amplitude noise from the signal generation AND the output buffer would need to be as good as the best dc implementation, as this noise will be directly transferred to the DHT, if present. Proper DC implementations achieve <<1mV of noise injection, so why settle for worse?
In practice, this means the power supply must be much quieter than the equivalent-performing dc solution, because the 40kHz signal source is going to multiply the problem.
Beware also of 32,768 or 40k “tuning-fork” crystals (i.e. the usual watch crystals) – these have very high sensitivity to supply noise, very-high impedance (ESR) and show large amounts of phase noise, which may also find its way into your output spectrum.
You also throw away the chance to see what high impedance dc drive can do for the sound, since the drive impedance must be LOW for any kind of AC.
On the practical side, the complexity is huge, and electrical efficiency a least as poor as linear DC.
I don’t see any motivation for this scheme.
Rod Coleman, August 2012