Decca 5501 Modifications for 136kHz

by Jim Moritz, M0BMU

I have now got my 5501 TX running on 136kHz - With a 60V supply, I get around 1200W into 50 . Previous Deccas that have been modified for 136k were the 128kHz variant, the tuning of which can be shifted by reducing the tank circuit capacitance a bit. However, my 6f variant originally tuned up on 85kHz, so required more work. As far as I know, it is the first of the 'Low band' variants to be so modified, and nobody has described what has to be done to the coils before, so I thought I would give some details.
The mods required can be summarised as:
" Modify tank circuit tuning capacitors
" Re-tune tank circuit inductors
" Provide power supplies, RF drive, keying etc.

Overview of Mods
The output side of the Decca unit consists of 3 identical full-bridge MOSFET switching PA modules. Each PA has a push-pull balanced output, which feeds a quasi-balanced series tuned tank circuit that effectively converts the square wave output of the PA into a sine wave. The tank circuit uses a large air-cored coil and a bank of polypropylene capacitors. Low frequency (5f, 6f) variants have a total tuning capacitance of 37.5nF made up of series/parallel 150n capacitors. High frequency variants (8f, 8.2f, 9f) have 100n capacitors giving 25nF total. Each of the 5 frequency variants has a different number of turns on the inductor. The tank circuit assembly also incorporates a "guard" circuit, which protects the PAs from short circuits. The outputs from the 3 tank circuits are combined using a simple transformer with 3 primaries.

The capacitors in the tank circuit (C5…C20) are 16 x 150nF, arranged as 8 parallel pairs connected in series as two groups of 4 pairs, giving a total tuning capacitance of 37.5nF. If the tank coil inductance was reduced to resonate at 136k, the loaded Q would be very low, and the filtering action of the series tuned circuit reduced. If the tank capacitance was reduced to get resonance, the loaded Q would be much higher, the tuning would be more critical and losses would be increased. In order to keep the loaded Q similar to its original value (about 6), it is necessary to reduce both the tank circuit capacitance and inductance. To avoid replacing all 48 tank circuit capacitors in the unit, I added four 100nF capacitors to each board, with a pair in parallel connected in series with each bank of capacitors, giving 300/300/300/300/200nF = 54.5nF in each bank, 27.3nF total. This is now similar to the 25nF in the high band units. I also replaced the 330nF capacitor C1, which in parallel with a 470nF C2 tunes the guard circuit link winding (mounted on the guard circuit heatsink assembly), with 100nF, making the total capacity 570nF. This is about in proportion to the change in tuning capacitance for the main tank circuit winding.

About half the turns on the tank coils were then removed to get peak output close to 137kHz. The link winding on the tank coils which connects to the guard circuit also needs to have turns removed to operate properly at the new frequency.

The output of the 3 PAs goes into a single combining transformer with 3 primaries and one secondary, with 1:1:1:5.2 turns ratios. With the design 72 load connected to the secondary, each PA sees a load of 7.9 . There is an on/off switch on each PA module which allows the unit to run with 1 or 2 modules; however, the use of a simple transformer rather than a hybrid means the load seen by the operating modules will depend on how many are operating. For example, running a single PA with the others switched off means it will only see 1/3 of the normal load impedance; with all 3 connected, the tuning of the tank circuits would interact. So that the tank circuits could be re-tuned individually under normal operating conditions, and to make life easier for future trouble shooting, I made a small 1:2.5 transformer, which when connected to a 50 dummy load gives the correct 7.9 load for a single PA. Having first got satisfactory tuning with this transformer, I then changed the ratio to 3:1 - changing the PA load to 5.6 , the same load as would be seen driving 50 s through the normal output transformer. I was then able to alter the guard circuit link winding to get satisfactory operation with the new load impedance. The benefits are that the supply voltage required for full 1200W output is reduced (from 67.5V with 72 load) to 60V, for which I had a suitable supply, and that no further matching is required for operation into a 50 load. Once the 3 modules had been individually tuned, the original output transformer was reconnected, and the complete unit worked at full power without further adjustment.

As far as I can see, the guard circuit acts as a current limiter. When the tank circuit current (determined by the load impedance) exceeds a certain ratio with the supply voltage (determined by the turns ratio and geometry of main winding and link windings), the voltage across the link winding forward biases the guard circuit diodes (D1…D4). This clamps the voltage across the link winding to the DC supply voltage and returns current to the supply. This effectively also clamps the voltage across the main winding of the tank coil, and due to the impedance transforming properties of the LC circuit, appears to the PA as a non-linear impedance in series with the load which increases rapidly when the tank circuit current exceeds a certain threshold. The rectified DC current from the guard circuit diodes is returned to the PA MOSFETs, and subtracts from the total supply current. If you monitor the 'guard', PA input, and supply currents using the front panel meter, as the load resistance is decreased, the guard and PA currents increase, but the difference between them (i.e. the supply current) decreases. With a dead short on the output, the PA current is increased by about 50%; after my unit was re-tuned:

With 50 load, 60V supply -
Total PA supply current for 3 PA's = 21.8A
Total guard current = 0.1A
PSU supply current = 21.8A

With short circuit load -
Total PA current = 34.9A
Total guard current = 28.3A
Supply current from PSU = 6.5A

An open circuit load just results in almost no supply current being drawn. After an hour running with the full supply voltage and a short circuit load, the PA modules were quite hot, but not dangerously so, so this is an impressively rugged design!

Modification Details:
Two of the new C's go in the vacant slots marked C21/C22, and there is enough room for the two others on the opposite corner of the board, once a couple of new holes have been drilled, and the PCB tracks cut to suit. C23/C24, if fitted, are not required. The capacitors used in the original Decca circuit are polypropylene film types, rated at 1250V. The new 100nF capacitors were all 1600V polypropylene radials (Evox-Rifa PHE428 series; RS components 240-5609). Other types of polypropylene film capacitors would also work well - but the type listed above fit into the existing PCB footprint nicely. C1 in the guard circuit is wired point to point and so is easily replaced. In total, 15 new capacitors are required - 4 for each tank circuit, plus one for each guard circuit.

The tank circuits were tuned up one at a time. The tank circuit coils on my unit (6f - about 85kHz) originally had 59 turns, plus some more wire bundled up for tuning adjustment. The link windings had 9 turns. With the modified capacitors, maximum output was at around 100kHz. To get maximum output at 137kHz, I reduced the main windings to 35 turns, with the final turn going in the opposite direction to allow slight adjustment. I initially got the resonance of each tank circuit close to the correct frequency by removing the PA module and connecting a signal generator and oscilloscope to the PA output terminals on the module socket. The output ends of the tank circuit are shorted together (but not to ground). Tuning the generator frequency gives a null at the series resonant frequency (see diagram)

Resonating the tank circuit at 137kHz

It is important to re-tune the coils in situ, because the screening is very close to the windings, and will affect the tuning. Once I had removed enough turns from each tank coil to get the frequency a couple of kHz below the wanted value, I put the PA module back in, and ran it up using the single output transformer on the frequency where maximum output occurred. To adjust the link winding, with about 30V HT supply, I monitored the guard current (which started off at an amp or so), and took turns off the link winding one at a time until the guard current was reduced almost to zero. This occurred when the link winding was reduced to 5 turns. Then I made the final adjustments to the main winding to peak as close as possible to 137kHz.

If repeating the process, I would leave an extra turn or two on the main winding of coils and trim as necessary, to allow for tolerances between units - it's easier to take the wire off than to put it back on! The link winding does not seem to be very critical. To remove the windings from the tank coil, you have to cut away the tape that holds the ends in place. When this is done, the wire spills off the end of the former rather easily, so when finished I stuck the ends in place with big globs of epoxy. G3LDO used a more elegant fixing; he drilled small holes in the formers and used plastic cable ties to secure the ends of the windings. The litz wire is of the type with 'self fluxing' insulation, so it was easy to tin the cut ends with a solder pot, by sticking the end of the litz wire in the molten solder for a minute or two until the enamel bubbles off. It is possible to modify the tank coils without completely removing them - this saves de-soldering their connections to the PA module sockets and threading the wires through, which might or might not be easier. If you remove the perforated bottom cover, the mounting bolts & spacers from the coils, unplug the blade terminals from the capacitor assemblies, and push the wires from the PAs through as far as they will go, the coils will come out of the chassis far enough to do the work.

The 'single module output' 3:1 transformer used for tuning the modules individually was wound on an E42 transformer core of 3C8 type ferrite (RS components 231-8785). Primary winding was 8t of plastic coated litz wire salvaged from some scrap Decca gear, about 1.5mm diameter. The secondary was 24t of 7 strands 0.3mm enamelled wire twisted together. Primary and secondary have to be insulated to withstand a few hundred volts of RF. It works fine to at least 450W. I believe G3GRO has the proper Racal-Decca test jig which does the same thing.

Once this was done for the three individual PA units, I re-connected them to the normal output transformer and ran up the complete transmitter with 50 load and 60V HT. The output was:
1240W at 135.7kHz
1240W at 137kHz
1180W at 137.8kHz
So perhaps I should reduce the inductances a fraction more. When tuned in this way, it did not seem to run any hotter than with 72 load. The supply current was about 22A, so efficiency is in the 90% region.

Other Mods
The input frequency is applied to the 3 PA driver boards via the RF input board, which contains a simple resistive splitter with 6 inputs. I disconnected 5 of the input leads and used the remaining one to feed in the 136kHz drive. The driver boards contain a Schmitt trigger squaring circuit, so as long as the input waveform is symmetrical and exceeds a minimum threshold, the actual level is not important. Several volts pk-pk sine or square wave will drive the input reliably. In my TX, I have used a TC4427 MOSFET gate driver IC to generate a 12V pk-pk square wave from a logic level input. The input is at 13.6MHz, and is divided by 100. On receive, the divider is disabled. You cannot control the output by varying the input level; it is "all or nothing" due to the Schmitt trigger inputs. An input signal near the threshold causes erratic operation, and in one case caused all the driver board fuses to blow.

The alarm board generates a status output for the rest of the Decca system to detect loss of output, etc., which I have just ignored. There is also a transmitter interrupt input, which disables the driver circuits when taken positive. I have disconnected this too. The monitor board meters various voltages and currents in the circuit, and does not require modification.

I have also added TX/RX changeover relays and a low-pass filter.

Power Supply
This is a hard to find item because of the power required. The original supply used in the Decca system was a nominal 67.5V, 20A from float-charged lead-acid batteries. G3LDO has used UPS batteries for his unit. G3XDV and G0MRF are using big switch-mode power supplies; I am using an ex-ATE power supply (Farnell H60/25). The output power can be controlled down to 1W or less by reducing the main HT supply voltage - the output RF voltage across the load is accurately proportional to the DC supply voltage over a very wide range, so a few volts can be used for tuning up. The voltage required to reach full output depends on the state of tune - if set up for the original 72 load, the original 67.5V will be required, but when operated into 50 , the requirement is reduced to 60V at about 22A

An auxiliary, fixed 27V supply is required by the PA drivers - the current required is about 0.8A

Keying
It is not possible to satisfactorily control the keying envelope by varying the input drive level to a class D PA such as the Decca in the same way as it is with a linear amplifier. To generate controlled rise and fall of the keyed signal to eliminate key clicks, it is necessary to control the DC supply to the PA. In my TX I have done this using what is effectively a high power, variable, linear series regulator. This sounds like a bad idea from the point of view of efficiency, but in fact the voltage drop across the regulator is small except at the keying transition, and overall efficiency is still around 80%. The average dissipation when sending CW of about 50W in the regulator for 1200W PEP RF out. It also works well for generating the half-sine wave modulation envelope for BPSK signals. At the moment, this circuit is something of a "work in progress" - I could write it up if there were sufficient interest.

In fact, most other Decca users have not bothered to do this, and key the TX drive signal at the input without any shaping. This does not seem to cause many problems with key clicks, probably due to the filtering effect of the high Q antennas used, combined with the smothering effect of the high noise levels on LF. Key clicks on 12wpm CW from G3XDV are just about audible at my QTH 11km away - I estimate that for a 1W ERP signal, it would be hard to detect key clicks a few 100Hz off tune more than 100km away. On QRSS, key clicks are not worth worrying about, because the keying transitions occur so infrequently that the clicks will be spaced far apart in time and not really a noticeable source of QRM. However, BPSK modes such as "Wolf" and "Coherent" do generate severe interference without envelope shaping

Miscellaneous Points
Re: Capacitors - The 1250V rated C's used, and the 1600V ones I added seem overrated - The complete bank of capacitors should withstand several kV, and the actual RF voltage is some hundreds of volts. But the manufacturers data for the 1600V units shows that they are rated at 630V RMS AC up to 6kHz, and then de-rated in inverse proportion to frequency. This implies that above 6kHz, current (per unit capacitance) is the limiting factor. With 630V RMS @ 6kHz applied to a 0.1u capacitor, current is 2.3A. When running into 50 , each PA sees 5.6 load, so at 400W out, the tank circuit current is 8.4A RMS, or 4.2A per capacitor - so the capacitors might actually be considered under-rated. However, the rating is for a 10degrees C internal temperature rise, so 4.2A ought to give roughly 33 degrees internal temp. rise, which is not unreasonable. The figures will be slightly different with the 150nF C's, but not hugely so. I think the ratings are probably 'worst case' - the capacitors in my unit don't seem to get much warmer than the surroundings, and the transmitter is not normally running continuously. But the view I took is that the new capacitors are somewhat over a pound each, which is much cheaper than IRF250's, and the fewer things that go bang, the less time is wasted, and the cheaper it is in the long run.

There is no special reason why the load impedance has to be 50ohm or 75 ohm - reducing the load impedance would allow the use of a lower voltage, higher current supply, but reducing the supply much below 60V would probably reduce efficiency also. All the components in the PA seem to be rated at 100V or more, so a somewhat higher supply voltage could also be used. The number of turns on the guard circuit link winding depends on the load impedance, so would have to be altered if the load impedance is changed.

G3XDV came up with a different method of resonating the tank circuits, but encountered some problems:
"The testing method was to remove the transmitter units and inject some RF from my existing Tx at the [PA module output terminals]. Putting a dummy load (a fan heater - measured at 40 ohms) on the output of the combining transformer, I then tuned my VFO for maximum RF current. So far, so good, but having found resonance at 109kHz [on an 8.2f unit], I cut off ten turns from the coil and re-terminated the Litz wire -a long process. My troubles then started as there was only a small change in frequency. After an hour of messing about unwinding (but not cutting) turns I found that the system resonated with no coil at all! Clearly something was wrong… I concluded that I needed to remove capacitors instead. Before doing this, I checked the unmodified coils for another puzzle. Whilst experimenting previously I found that my normally effective ferrite rod had almost no affect on the resonant frequency. Plainly something was wrong. I remembered reading that the amps had an output of only a few ohms, so I altered my RF source to the direct output from my BK amp (3 ohms)instead of after my step-up transformer. This produced quite different results: the ferrite moved the frequency by 10kHz or so, and the resonance was much more pronounced. Removing a few capacitors brought the units easily to resonance. So, it is essential that testing is done with a low impedance source. I wasted several hours with this one."

The method of finding resonance described previously does not suffer from this problem, and is insensitive to the generator output impedance.

Other Info
A simplified circuit diagram of the Decca PA, with some discussion on the guard circuit is at:- http://www.wireless.org.uk/guard.htm

G0MRF has posted some details of his Decca modifications at:-
http://www.g0mrf.freeserve.co.uk/decca.htm