Linear Amplifier Design notes

• Click here to download the 4CX1000A datasheet (PDF 448kB)

	The vertical axis of the graph is the grid voltage and the horizontal axis is the anode voltage. The solid lines plot contours of constant anode current and the dotted lines plot contours of constant screen current.
	The load line is determined by first selecting a point "A" on the graph which is the lowest peak voltage that the anode will be allowed to reach in a cycle of RF. Here the minimum anode voltage has been chosen to be 750V and the grid voltage -25V, these peak conditions result in a screen current just below the 50mA trip point.
	The ZSAC (Zero-Signal Anode-Current) recommended for the 4CX1000A is 250mA. This amplifier can use one of three possible transformer primary tappings (220V, 230V or 240V), resulting in anode supply voltages shown on the graph as points "B" 2.5kV, "C" 2.75kV and "D" 3.0kV.
	So for each supply voltage there is a loadline and thus an optimum anode load resistances (R=V/I) as shown below. A to B; (2500-750V)/(1.6 - 0.25A) = 1296 Ohms A to C; (2750-750V)/(1.6 - 0.25A) = 1481 Ohms A to D; (3000-750V)/(1.6 - 0.25A) = 1667 Ohms

• Click here to download an Excel spreadsheet for calculation of Pi- and Pi-L Tank values, coil inductances and dimensions (Excel 91kB)

Anode Choke testing

This photo shows the RF Parts RFC-4L 500uH anode RF choke mounted in a die-cast box to measure its impedance with a network analyser. It was mounted like this to see if any resonances would be introduced by capacitance between the windings and surrounding metalwork.

	The impedance plot of the RFC-4L choke from 1MHz to 100MHz. Parallel-resonance is at around 3MHz, at 30MHz it has a capacitance of just 4pF. There are no series resonances between 1 & 100MHz. A truly remarkable RFC.

Screen decoupling

As the anode current varies with signal, the electron flow from the cathode past the screen to the anode, varies the screen current. Current can flow into or out of the screen supply. The screen is biased to +325V by a shunt regulator which is capable of sourcing and sinking up to + and -60mA. Stiffly maintaining the screen voltage at 325V is an important factor in maintaining linearity and preventing distortion (splatter). Apart from the constant voltage requirements at low frequencies, at a syllabic rate, decoupling the screen at RF frequencies is also required to prevent any modulation of the screen voltage.

	This is a screenshot of the impedance seen by the screen from 1MHz to 30MHz. It is measured with the valve removed from the base and looking with a network analyser into the screen supply. Series-resonance, where there is a very low impedance to ground, occurs at around 14MHz, at all frequencies from 1 to 30 MHz the screen is decoupled by less than 5 Ohms.

The performance of the screen decoupling was optimised by using 35 turns of wire on a 10 Ohm 2 Watt metal-oxide resistor for the RF choke to isolate any resonances between the HF and LF capacitors on the screen supply.



Secondary Emission in second-hand "pulls"

The first valves I tried in this linear were two "pulls". When I first started testing the amplifier it became obvious that they had been taken out of service due to "secondary emission".

Secondary emission is where the screen gets hit by electrons from the cathode and is triggered into emitting more electrons than it gets hit by. This is a condition found in old valves where the screen has deteriorated and it results in somewhat "badly behaved" screen currents. If the secondary emission isn't too bad these valves can continue to give years of service as long as the screen supply is capable of good voltage regulation at the higher currents.

A third valve was eventually bought which was still in its manufacturer's sealed wrapping. It was a relief to find it behaved exactly as the constant-current curves in the data sheet indicated it should.



Intermodulation distortion, speech & 2-tone tests

Here's a screenshot of the transmit spectrum. It was made by speaking into the microphone normally for a minute and accumulating the trace with the spectrum analyser on peak hold. The lower trace is the barefoot FT1000MP, the upper trace is the output of the linear. It shows approximately a 10dB power increase with a clean spectrum.

This measurement was performed (in 2002) with and an Agilent ESA-L spectrum analyser, with a resolution bandwidth of 3kHz. This is what is seen by an SSB receiver with an IF filter bandwidth of 3kHz. However such a filter is too wide to show the amplifier distortion products when used with speech.

The intermodulation products of the amplifier were measured (April 2005) more thoroughly by driving the amplifier with two transmitters. When combining the outputs of two transmitters it's necessary to isolate them from each other to protect the receiver front-ends from damaging levels of transmitter power. This was achieved with a Wilkinson combiner using two quarter wave 70R coax cables. Also an Advantest R3361A spectrum analyser was used, which has a lower resolution bandwidth.

Screenshot of the coupler simulation using RFSim99.

Simulator results showing an expected 45dB of isolation.

The 100R balance resistor (only 5% tolerance).

The combiner isolation measured with a N2PK VNA using VNA4Win, showing 38dB of isolation.

The equipment set up to measure amplifier distortion.

The transmitter power levels were adjusted to provide 100W PEP to the amplifier, (shown here with the transmitters 50kHz apart). Two 25W carriers are combined to produce 100W PEP.

The spectrum of the combined signal of the two transmitters, 10kHz apart, into the amplifier. 100W PEP with 3rd-order intermodulation products 47.7dB below each tone, (-53.7 dB PEP).

The amplifier output spectrum, 1kW PEP, showing 3rd-order intermodulation products 32.2dB below each tone (-38.2 dB PEP).

The output spectrum of the FT1000MP (driven with music) measured with a 300Hz resolution bandwidth. +10dBm is equivalent to 100W (the analyser is preceded by 40dB of attenuation).

The output spectrum of the amplifier, (+20dBm is equivalent to 1kW), with the same music source as above, the spectrum analyser is in max-hold accumulating the transmission for over 11 minutes.

The input and output spectra overlaid on the same scale, showing spectrum widening 30dB below the peak.

Change from L-Pi to Pi-Tank

(May 2005) The amplifier had a fault where it tripped on excessive screen current on the LF bands, (80 & 160m). This was found to be caused by the L-Pi network, i.e., the small inductance between the anode and Ctune caused a parasitic oscillation when Ctune was set for the LF bands (>300pF). This was obviated by changing from an L-Pi to a Pi network. The 10m coil was adjusted by shorting turns with pennies.

Using pennies to short turns in the 10m coil of the Pi Tank.

The pi-tank matching range was measured by loading the anode with 1410 Ohms (3 x 470R resistors in series) with the 4CX1000A in situ and with the power turned off. The output impedance was measured with a N2PK VNA and the tune & load capacitors adjusted for an output impedance of 50 Ohms.

Testing the Pi Tank matching range with a N2PK VNA using VNA4Win.

Harmonic tests

(June 2005) Harmonics levels measured relative to a 1kW CW carrier.

160m, 2nd harmonic -45.3 dBc.

80m, 2nd harmonic -47.0 dBc.

40m, 2nd harmonic -58.5 dBc.

30m, 2nd harmonic -56.3 dBc.

20m, 2nd harmonic -62.9 dBc.

17m, 2nd harmonic -70.0 dBc, 3rd harmonic -64.9 dBc.

15m, 2nd harmonic below noise floor, 3rd harmonic -62.9 dBc.

12m, 2nd harmonic -45.3 dBc, 3rd harmonic -46.9 dBc.

10m, 2nd harmonic -54.25 dBc, 3rd harmonic -51.8 dBc.

Harmonic Level Investigation

(July 2005) The harmonics, on some bands, are much lower than expected, and I wanted to know why. On the bands where there are shorted mutually-coupled turns, the harmonics are attenuated much more than can be accounted for by a simple Pi tank output filter.

This Pi-tank is transforming 1400R to 50R, with a working Q of 12. So the Pi-tank filter alone should provide a 2nd harmonic attenuation of about 30dB. At 1kW the amplifier is running near class B with an anode current that is a half sine. So the resultant 2nd harmonic level can be reasonably expected to be about 35-45 dB down. This is roughly the case for 160, 80, 12 & 10m, but not for the bands in between.

These are the levels of second harmonic relative to 1kW on each band, and also shown is how many mutually-coupled turns of L4 are shorted out by the band-switch.

Band	2nd	Mutually-coupled shorted turns
160m	-45.3 dBc	no shorted mutual turns
80m	-47.0 dBc	no shorted mutual turns
40m	-58.5 dBc	7 shorted mutual turns
30m	-56.3 dBc	7 shorted mutual turns
20m	-62.9 dBc	10 shorted mutual turns
17m	-70.0 dBc	13 shorted mutual turns
15m	< noise	13 shorted mutual turns
12m	-45.3 dBc	no shorted mutual turns
10m	-54.2 dBc	no shorted mutual turns

After a helpful exchange of emails on the Amps Reflector it was realised that the section of the Pi-tank that gets shorted out by the band switch was acting as a trap to frequencies near some of the harmonics.

This test setup was used to measure the Pi-tank (with the amplifier switched off of course). The Pi Tank was set to 15m to investigate why the 2nd harmonic of 21MHz was so low.

And here it is, a 75dB notch on 15m’s 2nd harmonic! The notch can be adjusted over a wide frequency range by adjusting the load control. It’s just pure luck that the parasitic parallel resonance in the shorted section is a trap on 42MHz when the load cap matches the fundamental to 50R.