(foto: jc, art: Johan Röing, Lulu: herself)

okay, so today is going to be a straightforward identity-less empirical synthetic hybrid kind of authentic substance-less imaginary day… and i think it best to just dive in head first. that means not really, i guess? but along the way, some basic reality (i know fake is all the rage, but am willing to risk disappointing some people…) will be met with open arms and test equipment. below you see an example device under test (DUT) which will be subject to several arrangements

with a 10mV 1 KHz sine wave signal, i measure the gain of my open loop transconductance amplifier stage, which uses an E180F VHF frame grid pentode and a MOSFET “source follower” buffer. with the current source load adjusted with enough current to get 140 volts at the plate, and about 60 volts at the screen, i get a gain of:

that’s an actual measurement. and below is a frequency sweep of the DUT… so along with +71dB of gain, in the pass band (roughly 20 Hz to 3KHz) i get also a first order roll off from there, on up. that is quite a useful bit of gain for one stage! but not quite enough bandwidth for audio, where we would really want some stability margin on either side of 20Hz to 20KHz. what does that mean?

well, to begin with, the 10 mV signal becomes 35V RMS at the output… that’s a gain of 3500 (!).

and below you see a THD + noise level sweep

the distortion directly tracks signal level… which should make sense to anyone. and you can see that not much beyond 10 mV RMS input / 2% THD + noise (the graph’s X/Y coordinates), the distortion increase shifts into a steeper climb: clipping begins and saturation is not far ahead. the graph is limited tp a 20% X axis, but one can see the approach of no return. somewhere close above 20 mV input, it doesn’t matter how much more signal goes in… this circuit will not do anything more.

above is the transfer characteristic with a level sweep of 1 KHz signal… and it repeats the conclusions i made previously. there is a generally linear response until a bit more than 10mV in, followed by a rise in nonlinear gain, and the approach of saturation. note that 50mV input still hasn’t saturated the output… and this “soft clipping” behavior is a feature of the circut. a good question to as now is, “how is the recovery behavior following “soft clipping”?

above you see a typical inverting pentode “transconductance amplifier” in schematic form, almost exactly the same as the DUT at the start. i have been working with variations of this circuit since the mid 90s. a voltage in is converted to a varying current through the pentode, which is loaded by a fixed current source (the DN2540 and IXYS parts). the difference between the two is buffered by the high input impedance of the MOSFET which follows the first stage. the current difference between the two needs to go somewhere! where?! ordinarily, the two current sources (one variable and one fixed) would fight each other until the plate fastened itself either to B+ or ground. but, in this case, there is a regulating mechanism at the screen grid. DC feedback from the plate, through the direct coupled buffer stage cathode, corrects the plate voltage. all the gain available at DC is helping to stabilize 0 frequency… and there is quite a bit. the 1 meg resistor and 1 uF capacitor filter out audio frequencies.

above i have adjusted the components slightly to make it easier for the following changes… and the buffer is now one half of a 5687 dual triode. i want to show some of the subtleties possible with pentodes when it comes to voltage to current conversion, and maybe a bit the other way around, as well. first, lets just get into it.

above is an LT Spice frequency vs. gain sweep of the open loop circuit above. it closely matches the measured response of the practical circuit from the start. they are very close. right around 3KHz, it starts to roll off from +70dB of gain. what happens if the plate gets loaded down? let’s start with a megohm. R9 below is added between plate and ground.

the gain drops to +65dB and the bandwidth widens. -3dB is roughly 5 KHz

ok, now if we really load things down… 33k from the 6688 pentode plate, to ground. current flows to the “source” (we know the convention is backwards from reality, right?) from both the pentode AND the resistor. any variation in the current through the pentode will be shared with the resistor. note that the fixed current source trim has been adjusted to bring the plate voltage up to where it was before loading it with resistor… this extra current is the *difference*. the difference current, delta I, will vary in proportion to the transconductance of the pentode. and that slope will convert via Ohm’s Law: I X R = E. if you are using a “high slope” pentode, or tetrode (!), you will have lots of gain. the more the difference current winds up in the load, the less gain you get. and the less that ends up in the load, the more gain you get.

open loop, without any load at all, we have +71dB, with a bandwidth or 10 Hz to 3 KHz. with a load of 1 meg, we lost 5dB of gain, but got a bandwidth of 5 KHz. now, the current source has to “supply” 4 mA to the 33k resistor (R9) as well as 4 mA to the pentode, in order to maintain the 140 quiescent volts (means there’s no signal) at the plate. so in this case, there is equal current through both the load and the tube. 8 mA for both. the benefit of the fixed current source is that it will not vary. if the tube draws more, the resistor gets less… and the other way around.

below is a sweep of the loaded transconductance amp. the gain has dropped to +45dB and the bandwidth is now a healthy -3dB at 50 KHz.

what about the linearity? with the same 10 mV input signal, there is a gain of +45dB (x 178), or 1.78 VRMS. a look at the fft sweep below shows the 2nd harmonic at -60dB below the fundamental. lovely, and with NO error correction.

this is a feature of sharp cut-off pentodes, used in this way. the primary distortion mechanism is 2nd order and very simple. as level increases, so does harmonic distortion… in the triode-like gaussian distribution. if anyone tells you “pentodes produce too much odd order distortion”, remember this. because that statement is false.

but what about error correction? we have one single gain stage with a gain of 3162… (+70db) reducing the gain to 178 (+45dB) is -25dB of feedback. for one stage, that can do a lot of good without much risk of instability! and it isn’t even that much error correction.

above you see the same gm amp stage with feedback from the output returned to the grid. there are paralleled resistances so the feedback resistor (R2) is not going to be exactly the 178:1 ratio you might expect for a gain of 178… but the sweep says it all:

now the gain matches the loaded down gm amp stage… +45dB. and the bandwidth has increased to -3dB at 50 KHz, same as before! as for the linearity… the second harmonic sits at -65dB below fundamental. that is a small improvement. this is for the same 10 mV in, 1.78 volts RMS out….

okay so now for some subtlety… we know from the measurements of the actual device that we can swing nearly 40 volts RMS before hitting the wall. open loop that was with a bit more than 10 mV in. but with the loaded down or error corrected circuit, we have reduced the gain and increased the headroom. one should be able to get 200 mV input, with the circuit above, before running into the 40 RMS mark at the output. and the model says the second harmonic rises up to -40dB (1%) right there. i am not going to plot that because it will be more interesting to simply make a practical gain of 10 (+20dB) block that can do many jobs and has the known headroom of swinging up to 40 volts RMS. but keep in mind that we already have a circuit with a gain of +40dB for millivolt level signals will NEVER get close to clipping. it is why i prefer this for phono front ends. insane headroom against pops and clicks.

above is the +20dB gain block. with practical voltages shown… the frequency sweep and fft are below.

-3dB lies around 900 KHz and the gain of 10 is spot on.

with the same 200 mV in, the 2nd harmonic is -90 dB below fundamental. 2 volts out.

so, we have established a way to work with pentodes (and tetrodes) that offers some advantages and flexibility that is simple to work with. but this is only the beginning! next time there will be some non-linear fun! and phono will be a good place to start.

(below is an actual gain of 10 gm amp with measurements)

tested with an AP 525

-1dB at 70 KHz. i added some “Miller compensation” to the circuit to improve recovery… which it did. and that was 10 pF between grid and plate. that reduced the bandwidth a bit, so the response is now less than the model, or the uncompensated practical circuit. however, it is more reliable and better sounding to me!

THD speaks for itself… 1 volt RMS output.