DHT Filament Heaters

Experiments with filament heaters for 300B SE. To power the board, Hammond 266M12 transformer was uses (2 windings of 6.3@3A). For the test, KT88 (set at 5V) and 300B (5V) were used. The ripple voltage in this experiment was about 30-20 mV (rms) with 0.1 Ohm sensing resistor (R4), and 7mV with 0.2 Ohm R4. The rectified DC voltage in this configuration is 7.8V.

To see if better results can be obtained with more high quality components, the idea is to replace:

  1. LM385-1.2Z with LT1004CZ-1.2
  2. R1 with LM334Z and set it for 100-200 uA with 500 Ohm resistor (the bias voltage there is 68-72 mV for 27-60 degrees C)
  3. replace LM358 with LT1884

The schematics and the board can be downloaded from here:

Improved version (with suggestions (1) and (2))

Using LT1004CZ-1.2 and LM334Z with 470 Ohm bias resistor did not give noticeable improvements. The high 100Hz noise was due to poor traces on the board which are not capable of handling 2-3A of current properly. There was a difference in potential between each pair of diodes going to plus and minus points. By shortening all the points from diodes (see thick blue paths in the image) to capacitors the stray noisy potentials dropped about 10 times. The final noise at the heater contacts is about 0.6-0.4mVrms (see the screenshot with x1000 amplified oscilloscope input using Philips PM5171). The spectrum is totally flat at the level of -76dB. R4 is still 0.2 Ohm. Using 0.1 Ohm leads to 2 times higher noise.

The results look quite decent. For comparison, while using CLC filter from 6.3V one can expect the following results (see the original post here:
  • 30,000uf no choke 5.42V, 190mV p-p ripple, saw tooth wave patter (not good)
  • 2. 10,000uf, 3mH, 10,000uf 5.8V, 50mV ripple, normal sine wave
  • 3. 10,000uf, 6mH, 10,000uf 5.4V, 25mV ripple, normal sine wave (better)
  • 4. 20,000uf, 6mH, 10,000uf 5.2V, 7mV ripple, normal sine wave (best so far)
  • 5. 10,000uf, 3mH, 10,000uf, 3mH, 10,000uf same as #4

Improved version (even thicker paths and filter for the voltage regulator)

The version above had about 0.4-0.5 mVrms at the output (5V, 1.2A). Extra paths around diodes and capacitors were added, which reduced the voltage drops between the diode connections and capacitors from 0.5mV to 0.3mV. The trick (described in the LT1004 datasheet) with 22 Ohm resistor and large capacitor on the voltage regulator reduced the output noise from 0.4 to 0.2 mVrms. All this was done with the 0.2 Ohm R4. Using 0.3 Ohm R4 produces output with 0.12 mVrms. One more experiment is to try better opamps (LT1884 instead of LM358).

Further experiments

Replacing LM358 directly with LT1884 caused no improvements but rather strange behavior: With R4 < 0.3 Ohm, there was no voltage regulation at all. With 0.3 Ohm, everything worked but the noise Vrms became 200uV instead of 120uV. With R4=0.5 Ohm, the regulation was there, but the Vrms was about 200-300mV with strong sine oscillations at 80 kHz. The problem with the LT1884 might be that according to specs, it requires Vcm (input voltage range) [Vee+1V, Vcc-1.V], which means that 0.3-0.5V on R4 are not enough. At the same time, LM358 works in the range of [(V-), (V+)-1.5], so the bottom part with R4 is fine, BUT, the “top part” with the 1.2V voltage regulator might be not optimal. Using LT1004 with 2.5V reference might be a better idea.

Back to LM358, changing R4 from 0.3 to 0.5 Ohm, lowered the noise Vrms from 120uV to 80uV.

With LM358 and R4=0.3, removing the 1800uF cap increased the noise Vrms from 120uV to 200uV. Reducing that capacitor to 330uF left Vrms at 120uV. With 1800uF cap the start of voltage regulation was always delayed for 16-20 sec.

The voltage regulation of the hole module is stable at 6.3V DC as well (with a 6.3V AC transformer) which means that the MOSFET is comfortable with Vds of 1-2V (7.7V DC as input, shared with 5V tube, 0.5V R4 and the rest is for MOSFET).

Final schematics (for heating of 300B)

  • trial with LT1004 2.5V did not improve anything
  • the 1800uF capacitor was reduced to 470uF. With that RC filter, the voltage on the filament increases slowly during the startup, within 3-5 sec.


In general, looks like a simple and decent regulator. The described by Kees Soeters PCB and its reimplementation will definitely should be redesigned with better handling of higher currents. The comparison with Rod Coleman’s regulator and Tentlabs modules would be interesting.

Update: after searching a bit, the following evaluation by DHT Rob was found, which is very comparable to the image above, meaning that the Vrms are almost the same, in the order of 100uV. Apparently very simple and cheap schematics can perform very decent compared to the most expensive models.

scales:  0.2mSec, 0.1mV/Div