Testing the Stage1

I put an oscilloscope capacitive probe on the input collector.

The stage1 works basically OK except is behaves strange. The measurement is on collector of the left (input) transistor. When the input signal strength is slowly increased, the output signal strengt jumps abruptly from 0.8V to 1.4V.

I set the signal strength exactly into the critical zone. Noticed moving a hand around causes the waves breath or jump periodically with distance of the hand from the circuit!

I took a large piece of tin (54x36cm) and moved away from the circuit. The phenomenon repeated every 17.5cm which corresponds to oscillation frequency of 860 MHz.

Now I tried to touch gold plated pin held held in a Kelly forceps against various parts of the circuit when the signal strength was just the critical band. Sensitive: input collector, both bases, both emitters, ground close to the ground clip.

Now I set the amplitude just above the jump. Sensitive points: probe ground near probe tip, both collectors.

Now tried which points when touched prevend this jump phenomenon: input base

One cause could be TMD resistor inductance (spiral groove) 6k8 in the base. The amplifier was not designed for that. I will disassemble the resistor to see what's inside. I could replace it with a SMD resistor.

The resistor has a spiral trace of 6 turns with thin gap between turns.

It usually takes a LC oscillator with low damping to get an oscillation. Where could one be? The intended capacitors in the circuit are too big for 860 MHz. It could be: protective diode 2pF with 10nH - transistor parasitics 5nH and 6nH - that makes 776 MHz. The probe input capacitance 16pF and inductance est. 150nH will cause the probe to behave as 150nH into the ground. This could combine with the Ccb 0.4pF then go through Ceb emitters and into the ground through the blocked base. This would give 650MHz. This is the first scenario from the Why Circuits Oscillate article!

If it's the probe it should be sensitive to deformations of the ground lead. It should also cease when the collector resistor is split into two and measured in between.

If it's the protective diodes it should show on a stability simulation of the stage 1.

I twisted the probe ground lead and it still jumps. The period of the reflections still seems to be the same - 15cm.

I disconnected the protection diodes and the jump ceased! I could try to damp them. That would require 140 Ohm in series. Parallel is not possible. 140 Ohm is too much.

What about 2x3 diodes coupled through a capacitor to the output of the previous stage to limit the signal to 4.2Vpp? After soldering these with a 100nF TMD capacitor and disconnecting stage1 I noticed the output signal strength abruptly drops from 3.5Vpp to 0. Isn't this some oscillation again?

I disconnected the signal limiter again and the same still hapens. I am desoldering the tin cover from the first two partitions. The signal disappears exactly in the moment when the photodiode voltage reaches zero. It is caused by DC saturation of the photodiode.

Now the diode limiter is connected again and the jump is away. Just at 2.5Vpp input signal there is a damped oscillation starting at each edge, amplitude small but visible as jagging very well. 3 periods have 27ns, the frequency is 111 MHz. It doesn't react to twisting of the ground leads and leaks into the power supply - starts at 22mVpp there. Could it be the resonance of the TMD 100nF capacitor blocking the base? Touching both bases or input with a golden pin in a Kelly forceps decreases the amplitude and frequency. It can't be a self-resonance of the 100nF because that resonates at 5MHz.

These artifacts also appear exactly when the amplitude stops growing - when the signal is fully limited. Some overshoots can also be seen in the transient response simulation. Let's pronounce this the limitting artifacts. The first stage doesn't have it easy since it's running in an asymmetric mode.

Switcher noise

There's a 60mVpp noise from the switcher on each differential output. The images are mutually inverted that means it doesn't leak into the output of the differential stage but into the input.

Band limited display

The 470 Ohm and 16pF input of the probe create a 21MHz RC lowpass which is less than the intended lowpass of the stage. I need to use a resistive probe.

Garbage on 1MHz square

When I send 1MHz square into the transmitter and set the amplitude on the input collector to 680mVpp, then a garbage appears in the middle of the bottom roof. When I move the probe to the non-input collector it is not there and the signal is only 230mVpp there. If I increase to 800mV then it;s there too, on the top roof.

On the input collector it reacts to chages in the shape of the probe cable loop, and to a lesser extent also touching the probe cable. This doesn't happen on the non-input collector.

With the capacitive probe it happens as well, but in both collectors on the bottom of the waveform. In both cases it's very sensitive to the movement of the ground lead of the cap probe.

One candidate could be the inductance of the TMD 100nF blocking the base. After replacing with SMD, the phenomenon almost completely disappeared!

What remained is a small notch and deformation of the bottom of the 1MHz wave on the non-input collector which reacts to touching the probe wire of the resistive probe. Moving a ferrite on the probe cable produces periodic behaviour with period of 10cm. That's wavelength of 20cm or frequency 1.5GHz. Happens at medium to high signal strengths, no signs in clipped state.

I did 4 marks on the cable and their distance is: 10.0, 10.6, 10.6cm.

If we estimate the probe inductance to 10nH (2 wide traces close together 20mm long) and capacitance to 1pF (no idea really how to estimate), then we get 1.6 GHz. It could be caused by the probe. Does a capacitive probe do the same?

With capacitive it's the same, about 13cm, not possible to measure precisely this time. Isn't caused by a particular probe.

I tried desoldering the protective diodes. Didn't help.

I replaced the input collector TMD 470R with SMD 470R. Didn't help.

Now I replaced the non-input collector resistor from TMD to SMD 560 (I didn't have 470). Also didn't help.

Now I will try an SMD 100nF between the two collector resistors. To avoid resonance I have to separate it from the other 100nF with a 1.5 Ohm resistor at least (critical damping). I will use 3.3 Ohm. This didn't help either.

Since it happens only when the probe is put into the positive output it could be a feedback through air. When the positive output is touched by hand, exactly the same oscillation happens in the same place.

Improving the probe ground inductance improved the problem but didn't remove it.

I will solder a piece of tin in an attempt to shield the input from the positive output. I soldered a rather large piece that covered the whole part with input components - interstage capacitor, previous stage resistor, the hole to the previous stage. Didn't help at all!

Now I remade the whole stage1 with closer spacing and less inductance in the emitter degenerator. It actually oscillates even much worse. Still 1.5 GHz.

Now I got an idea to symmetrize the input by using a double wire going across the previous working resistor and omitting the 100nF.

That works for linear mode but starts failing in limitted mode, but only unilaterally. That suggests the problem could be an asymmetric input impedance. I could try doubling the 220 Ohm working resistor of the previous stage.

That works on 10 and 5 MHz limited and unlimited. Just in 1MHz there is garbage on the bottom when the probe is in the inverting output and slight on the top when in the noninverting one.

Since the amplifier is now electrically symmetrical, it should also oscillate symmetrically. This suggest an asymmetric feedback or maybe it's cause by asymmetric output loading? I will manufacture another probe and connect it into the other oscilloscope channel.

First I swapped the 100nF and the 820R (replaced with 680R) to reduce stray electric field from the probe. This has improved the problem.

With the other probe the problem didn't disappear is the same. That shows unilateral loading is not the cause. I found when the input twisted wire is pressed down to the ground the amplifier can be actually made to amplify properly.

Turning 90 degrees

I shortened the input leads dramatically by turning the whole amplifier 90 degrees. That didn't help.

Wen I estimate calculate the ground lead inductance - 20nH - that has 188j Ohm reactance! The lead works basically like a bare wire plugged into the output.

I changed the probe construction and the connector to use 2 leads with estimated inductance 10nH each that's only 5nH. That's 47j Ohm reatance still a lot!

But the wire doesn't seem to be a problem. It oscillates even when I plug the "electronics" from the old probe, without the cable, into the pins of the inverting output. If I turn it so the ground part (which runs on the most of the PCB) is in the signal pin, it goes totally crazy. When I plug only the ground part into the output and the signal part is floating, it goes also crazy (not so totally). If I plug only the signal part into the output, it oscillates, but with a reduced tendency.

If I plug a bare pin into the output, it also oscillates (not very readily). Without any pin it doesn't oscillate at all. The same with 2 pins into signal and ground.

This suggests the problem might be conveyed electrostatically.

1pF capacitor

A long-legged 1pF capacitor causes oscillations like hell. Only in output as well. In the ground not.

1pF with short pins makes oscillation like crazy too. Into output only it makes barely noticeable ones.

Shield

Now my theory is that a shield is missing and the radiation crosses from the output to the input. I made a shielding partition across the transistors and it didn't help. Now the favourite theory is that the additional things cause capacitive loading of the output with a small capacity and this somehow causes instability.

I decided to remove the capacitive loading by putting a 820R directly on the collector output in series with the measurement connector on the side where insertion of the probe caused oscillation. Now it doesn't oscillate on 1, 5, 10 MHz!

Even plugging the bare PCB one other way and also only one end didn't cause any sign of oscillation! Adding another 820R into the othe output is also OK.

Now I will replace the 470R's with 390's and 82's to make a direct interface for the cable and short the 680R inside the probes. This is no problem, runs nicely and is not sensitive to random touching of the circuit with finger.

Now adding the diode limitter (2x3 diodes + 100nF) back. That works also fine.

Now time to remove the impedance balancer from the input, which creates extra noise. Also seems to make some extra glitches on the edges. Unfortunately, this oscillates! The contribution of the resistor to the noise is something on the order of 0.6nA. The stage1 noise is 278uA and that corresponds to 1.5nA at the input. The resistor increases the noise by 116uA. I am not calculating the effects on bandwidth which will result in a need to redesign the whole stage1.

Therefore I am deciding to put the input impedance balancer back to get a nice clean stable running amplifier!

Power noise

The 12V line at the switcher now contains switching noise. The power consumption of the supply is 7mV (measured) and doesn't vary with photodiode consumption. This has changed after I redone the regulation to use a transistor and be able to supply 6mA.

For a filter I will employ 10uH and 100uF with 100nF in parallel. This has a resonance at 5kHz and element impedance 0.31 Ohm. Critical series damping is 0.62 Ohm so I will use the 1/2 Ohm resistor. The electrolytic has some internal resistance anyway. The voltage drop will be 3.5mV, that's excellent.

The 12V (the same wire as from the battery) now contains ringing 20MHz which doesn't disappear even when the probe (1:10-1:1 set to 1:1) is shorted to the ground or removed completely from the device. The noise may be transmitted through the air.

So I shielded the power supply and now the 12V contains only 400uV triangular ripple from the power source. However I lead out a sync optical signal for the oscilloscope using 680 Ohm resistor on one of the output. This makes approx. 11mApp rectangular power draw signal whose remains may get through the power filter. What I try is to add another LED (ordinary not high efficiency) on the other leg. The result is: reduction to 200uV only and the frequency of the ripple is 100kHz instead of 50kHz. horting the photodiode changes the ripple shape and increases it to 400uV. The PSU gets overloaded as can see on distorted waveform from the monitoring LED.

I can also see about 200uV noise, presumably from the battery (scope BW 5MHz). This could be potentially fixed connecting the battery through an old transformer and an electrolytic capacitor.

200uv 100kHz remnant

Apparently the power filter needs to be improved to get invisible ripple. The current drawn by the circuit is distributed between 100uF and a branch with 10uH and 100uF in series. 100uF has -0.016j Ohm and the coil 6.28j Ohm. Therefore the current through the branch is reduced by reducing the capacitor closer to the power supply. The voltage on the capacitor farther away from the supply is reduced by making the capacitor bigger. Increasing the coil has the same effect.

Increasing the coil increases the damping resistance, increasing the cap decreases it. I will increase the capacitor because I don't have to increase the damping resistor which causes voltage drop for the supply.

Adding 1000uF parallel to the main supply eliminates the ripple. Not tested with shorted PIN.

Adding 470uF parallel to the main supply decreases the ripple to barely noticeable. With shorted PIN it's still about 200-300uV. I wanted to put 470uF instead of 100uF into the PSU utself, but I will put 1000uF. The inrush will be at least somewhat attenuated by the resistor and the coil.

2 reasons for being conservative here: the ESR of the caps is unpredictable and I don't want the device to radiate stray frequency over the power line.

There is still barely noticeable ripple. about 100uV, with the PIN shorted about 200uV. I will increase the coil to 47uH now. I will not increase further if not necessary because I am afraid that the resulting coil would be too big , difficult to get, or expensive. The damping resistor should be 0.77 Ohm so I will put 1Ohm there. The effect of this resistor on the output impedance is only 6 Ohm.

With PIN unloaded nothing can be seen. With PIN shorted, the ripple is barely noticeable. Therefore I will increase the other 470uF capacitor to 1000uF. Now the ripple cannot be seen even with the PIN shorted however a barely noticeable notches (high frequency) can be seen under the edges. They disappear when the PIN is unloaded. I tried increasing the coil to 100uH but it didn't help. However I will keep the coil 100uH because I am lazy to put it back and also it will create a reserve for case of less quality electrolytic capacitors.

Now how is the HF leaking? Isn't it thgough the touching grounds? No, when I lift the photosensor it doesn't cease. Isn't it oscilloscope crosstalk? No, when I put the photosensor away, it doesn't diminish. One possible route is lousy shielding (soldered only in 4 points). I am trying to solder the shielding completely around the edge. Now it has disappeared completely! :)

This is the same picture when the 1000uF electrolytic is disconnected to show the ripple: