Eliminator was built for this "Reknown Special 3", but suits most battery valve sets.

Need for a battery eliminator.
The first generation of radio valves were all directly heated. That is to say, the filament, or heater, functioned as the cathode. As a result, AC operation from the power mains was not possible because of the amplified signal being modulated at the mains frequency. Furthermore, in the 1920's, mains power was not as widely available as in later years. So, radios ran from batteries. For the heater (A) supply, a lead acid accumulator was used. High tension for the plates (B) usually came from dry cells, although a battery of small lead acid cells was sometimes used. The grid bias supply (C) could be provided by dry cells which would last their shelf life, as no current is drawn. Typical voltages might be 4V for the heaters, -9V for the grid bias, and 135V for the high tension, although this depends on the valves.
Of course it can be imagined that having to take a lead acid battery to be recharged, and the cost of replacing a 135V dry battery is not without disadvantages.
Once power mains started to appear, the natural response was to want to run the radio from them at a next to nothing cost. And, so battery eliminators appeared on the market as an extremely popular accessory. A typical example is the Philips 3003. Obtaining the B+ and C- voltages was quite easy using conventional valve rectifiers and paper filter capacitors of about 4uF each. The problem was the A+ supply. At the low voltage and high currents required; e.g. 4V @ 200mA, ordinary valve rectifiers were unsuitable. Where attempts were made, metal rectifiers could be used. However, the real problem was filtering with the available capacitors being totally inadequate. Crude electrolytic types were sometimes used, but were not completely satisfactory. In fact, it was a virtue that the low frequency audio response of a 1920's receiver was so poor, enabling a large amount of hum to go unnoticed. One  successful scheme was to retain the filament battery and trickle charge it via a Tungar or metal rectifier. Because of these difficulties, most eliminators only provided B+ and C- supplies.
Once indirectly heated valves became available in the late 1920's, battery types were relegated to portables or for sets used in areas with no power mains.

The Reknown Special 3.
This set is a Wireless Weekly design from 1929 and is very typical of the time. It is a three valve regenerative set; one detector, and two audio stages, all using Philips triodes. The speaker is a Baby Sterling horn type. Correctly speaking, this predates the radio by a few years, but looks the part. Since I have had this set, I've powered it from a Philips 3003 battery eliminator for the B and C supplies, and an equally ancient Philips trickle charger using a mercury rectifier and baretter. This powers the valve heaters by a solid state regulator, providing 4V with no hum. While this set up worked perfectly well, it was cumbersome with three boxes stacked on the floor and untidy wiring between them. I wanted something more compact and neater.

The new Eliminator.

Much more tidy and compact, this proves all the voltages required from the 12V house lighting plant.

As with a lot of things I build now, I always consider operation off the 12V supply reticulated throughout the house, thus eliminating running costs. The new eliminator would be ideal for this. I decided I'd rebuild one of the vibrator power supplies I'd had lurking around for years. The type of vibrator power supply I'm talking about was the kind used to convert dry battery sets used in rural areas to run from a 6V accumulator. Typically, an output of 135V at about 20mA is provided. The valves are 2V heater types which are rewired in series parallel to run from 6V. Sometimes the order in which the filaments are connected will automatically provide the correct grid bias, but as a dry battery would last many years providing this, it was a minor point.
A synchronous vibrator is used in such supplies to avoid the heater current of a valve rectifier. Units which do supply C- use a split reed vibrator, such as the V5211. This allows the C+ to be subtracted from the B+ supply via the isolated reed on the secondary side (the one that does the rectifying).
The supply I decided to adapt and rebuild was a typical late 1940's design containing a V5124 synchronous vibrator and a transformer, filter choke, and capacitors providing about 140V at 20mA.
Connection to the unit was via a UX 5 pin socket at the side of the box. It appeared to be of backyard manufacture, but commercial enough to have a serial number.
My rebuild would provide B+ as well as C- and A+ supplies.

The Design.
First thing was to adapt it for 12V. Here a resistor can be used, but it is essential that the load remains constant, otherwise the vibrator and transformer will be exposed to excessive voltage. One way I considered was to simply connect a VR150 regulator valve across the output. Regardless of the radio load, the primary current would remain the same. Alternatively, a high power 6.2V zener diode could be connected to the 6V side of the inverter. What I decided on solved the problem very neatly, as well as providing multiple B+ voltage taps, all of which were regulated. One characteristic with regenerative receivers is that they are voltage sensitive. Problems can arise here with unregulated power supplies because as B+ current changes so does B+ voltage. So, fluctuations on the mains supply can send a critically adjusted receiver into oscillation. WIth some receivers, such as the Reknown Special 3 that use filament rheostats, adjusting these changes B+ current, and thus voltage, again causing the problem.
The C- supply was to come from the transformer secondary rectified by the synchronous vibrator, and divided down to provide a variable output.
A+ was to come from the 12V supply via a resistor and high power zener diode.

Final circuit of the eliminator.

B+ Supply.
Initially, I simply re used the original 6 to 140V supply, but stabilised the output with a series of 10W zener diodes. These provide a constant 20mA load on the inverter, regardless of what current the radio draws. The zener diodes were chosen to provide all the likely voltages of typical battery radios. 30 and 45V are usually used for the detector stage, with higher voltages for the audio amplifier and output.
The vibrator is a synchronous type to avoid using a rectifier valve. The two original 8uF electrolytics tested OK, but the buffer capacitors were changed as a matter of course, as leaky ones will ruin the vibrator, and possibly the transformer. I mounted the zener diodes on a piece of PCB with the laminate cut into rectangles. This provided an easy and neat form of connection while also providing isolation between the diodes.

The zener diode panel is visible here. It provides a constant load to the vibrator and transformer, as well as ensuring all B+ voltages are regulated. The inverter's dropper resistor is also visible.

A 6R 20W resistor drops the 12V for the vibrator and transformer. As can be guessed, current draw is 1A. The existing filter was left in situ, although a smaller choke, made by Van Ruyten, replaced the original much larger one. This was necessary in view of the space taken by the zener diode panel. In order to tweak the load current, a 470R resistor was connected in series with the filter choke. This was necessary to limit the secondary current to 20mA, while allowing the input to remain at 6V. All worked perfectly, but when it came to the C- supply the vibrator circuit had to be modified.
I had imagined that by various kinds of decoupling circuit that it might be able to pick up a low current negative voltage at the secondary vibrator contacts which I thought might be present when they open.
Alas, this was not to be so. I considered a split reed vibrator, but I did not have a UX 7 pin socket that would replace the original 6 pin one without modification. I thought of a valve to rectify the AC from the transformer, but the problem here was the set would be deprived of bias until such a valve warmed up.
So, I took the easy way out and used silicon diodes for both B+ and C-. It was not possible to retain the synchronous rectification provided by the vibrator, because to do so requires the secondary winding centre tap to be the positive output. This prevents a simple rectifier being used to obtain a negative voltage from the outer connections. I used some late 1950's diodes. The vibrator was rewired in non synchronous mode by connecting the primary and secondary contacts in parallel.

C- Supply.
A simple half wave rectifier using a 1N3194 diode is used to rectify the 140V AC present at the transformer secondary. This is divided down and regulated by an 82K resistor and 18V zener diode. With a 25K pot across the zener diode, any C- voltage from 0 to -18V can be obtained.

This side shows the filter choke and transformer. One of the 8uF electrolytics is visible.

A+ Supply.
For the Reknown Special 3, the heater requirements are 4V at about 180mA. Again, the easiest way to get this is with a resistor. Because the A current can vary due the rheostats, it is possible the valves could be damaged under some conditions. If the rheostats are adjusted to drop filament voltage, the A+ would actually rise and this would be exposed to the other valve filaments in the circuit. Also, it must be realised that the house supply can vary anywhere between 11 and 15V. Given the fragile filaments of the valves, plus the fact the gain would change with filament voltage variation, it is obvious the A+ must be regulated. Here, a 4.7V 10W zener diode provides that function. The 20R resistor is the highest practical value that will maintain 4.7V on the output with the supply input at 11V. The 20W rating is overkill, but such a resistor was easily mounted on the chassis. Incidentally, if reverse polarity is fed into the unit, the zener diode will be fully biassed on and there will be only about 700mV on the A supply. No damage will occur. The inverter, now with the vibrator wired in non synchronous mode, does not care about input polarity. However, with polarised plugs and sockets, this scenario is extremely unlikely.

To the left is the A+ dropper resistor and zener diodes. In the middle are the RF filtering components, including the circular RFC. At the right is the vibrator and rectifier circuitry.

The use of a 4.7V zener diode to create a 4V supply might cause question. In reality, zener diodes have tolerances, and are not perfect regulators. The actual voltage will depend on the current to a small degree. As it happens, with the current from the 20R resistor, the voltage developed across the particular zener diode used is just over 4V. By the time it reaches the valve filaments through the wiring and rheostats, it is so close to 4V it doesn't matter. The low internal resistance of the zener diode means that any further filtering was not found to be necessary, despite the directly heated filaments.

Back of unit shows the terminals and bias control pot. This has a cap over it to prevent accidental adjustment. Instead of multiple boxes, the wiring harness from the radio converges just on this one box.

Performance.
The new battery eliminator works as well as it was hoped it would. No RFI problems are evident, and it is mostly immune to supply voltage variations. It is certainly no worse than when the unregulated mains supply was used. With the current draw of only around 1A, it is perfectly satisfactory to run it all day if desired without running down the house batteries. My Reknown Special 3 is probably the only 1920's coffin set in the world running on solar power!