Electronics - Designing with discrete semiconductors

Created; 12/10/2015, Changed; 04/03/2024, 03/03/2024

Previous page; Electronics - Noise Cancellation

There are many different transistors, and a lot of those are the same dice (component) but selected for different parameters and available in a range of part numbers.  Some types such as Bipolar Junction Transistors with symmetrical Emitter and Collector junctions providing bidirectional high voltage blocking but do not appear to be available now.  Evidently they are made and PNP power transistor are within many low voltage drop regulator IC's to withstand high (automotive standard) reverse voltage.  Designing with transistors can give low-cost solutions or solutions that fit in to small or odd shaped spaces.  You should so more modelling than you would with an IC and there are tools, but I use pencil, paper and calculator or a sheet.

Some of the types not commonly available now are;

Some solutions using transistors over a wide temperature and voltage range so require extensive modelling using specific transistor data as opposed to using general rule of thumb rules.  For example, much of the lighting in automobiles must be of a specific colour and intensity over a very wide temperature range and must also fail unambiguously if an LED goes open circuit but not dim, so causing a partial failure.  The circuit fit in a small space.  These low cost, robust applications are the most demanding to model. 

High voltage driver for particle deflection vacuum tube applications; 

The tube is 3BP1, made in USSR. 

The CRT is not in this data book.

The deflection transistor circuit in this Practical Wireless (1973) constructional project oscilloscope used MJE340 transistors, which are called medium power transistors, and they are quite slow, <10KHz.  I replaced them with TO5 package TV video driver transistors that were better giving >500KHz but excessive power consumption, but I learnt later that some small E-line Ferranti transistors such as ZTX458 certainly would have done a good job and without the excessive power consumption of my design.  I used lowish value resistors as collector pull up to improve the speed.  The driver would have needed to be moved to just behind the Cathode Ray Tube base to keep the board tracks and the stray capacitance down that would otherwise be significant.  The bootstrap BJT input amplifier is a particularly slow function block, and a FET input I changed it for would be a big improvement as well.




The Valve Museum has more circuit examples of this age or older.  Notice that they can have wires crossing, and circuit function spread about, making the circuits difficult to follow and mistakes; http://www.r-type.org/

The Cossor oscilloscope, manual pictured was probably purchased after 1967, 20MHz, chopped dual trace, with wound delay so that the trigger signal can be seen. 

The Y plate amplifier uses a 50V supply, a 10th of the much older CRT in my homemade oscilloscope.  The circuit has many trimmers and compensation components.  It also uses nuvistors rather than FET transistors where high impedances are required, and uses common-base for the plate drivers.

The schematic (circuit diagram) is, as is typical of the time, somewhat unclear because functions are split, such as the second and subsequent differential amplifier stages split.  This is typical of circuits of the time drawn up by a drawing office.  Availability of CAD tools 20 years later resulted in designers finishing off the circuit diagrams more clearly, similarly if they did the PCB design then the completed design works very much better.  The GE Thyristor book on a previous page is by comparison clearer.

I had considered using a valve amplifier output stage - there was a lot that I did not know at the time, particularly understanding the reading of data in data books.  I had purchased this 1970 edition data book and now see that a pentode valve would have had a tiny fraction of the miller capacitance that any transistor has (<<0.1pF) so the power demanded would have been resolved as well because it would not be necessary to overcome the components capacitance provided I placed the driver very close to the CRT base.  There would be heater power of 2W per valve.  Manufactures data books, even old ones, are invaluable to people starting a hobby.  A pentode valve has two screen grids between the control grid and the anode and reduce ga capacitance by 20-fold I observe from the data book compared to a triode valve's 1-5pF.  So pentode valves would still be a better option in this case where the CRT has such low sensitivity. 

I have criticised the circuit below, then changed the circuit showing some of the steps.

I have introduced snubber networks (R+C) that dampen switching circuits previously, and I have also introduced gate drivers.  The circuit below has such damping, but it is not so plainly shown, but the transistor stage has a limited gain (proportional) and capacitance (integral) those parameters are also dealt with within the op-amp in order to make it unconditionally stable (for gain of 1 or more). 

The circuit and its application will require very good electromagnetic and electrostatic metal screening, as well as distance kept between it and anything that could interfere, such as a power supply.  That is the same care issues that I mentioned for the oscilloscope.

The bandwidth limit occurs when the signal has dropped by 3dB from its regular amplitude, but I have used a point roughly when the signal is just beginning to drop and therefore gives a little margin for manufacturing tolerance, on this page. 

The major speed limiting factors are the transistors miller capacitance and the high value collector resistors.  Driving the circuit from a low impedance - C1 does this.

The T resistor network configuration in the emitters works out more useful than a pie configuration.

This differential amplifier works best with a differential input signal or a small signal from a low impedance driver. 

If the CRT's capacitance had been lower than the driving transistors' capacitance, this circuit would be about as good as can be done.

AL-0007-01B Conventional high voltage differential amplifier. (OrCAD capture)

Notes - Transistor selection; 


MMBT6520L PNP 350V 500mA; Vceo at 1mA, Vcbo at 100uA, Ceb=100pF and Ccbo=6pF 


FJT44, NPN, 500Vcbo 100uA, 400Vceo 1mA, 300mA, Ccbo=7pF,

 Diodes Ltd

MMSTA92, PNP, 100mA, -300 Vceo 1mA, -300Vbo 100uA, Ccb = 6pF 

DXTP560BP5, PNP, -500Vceo 10mA, -500Vcbo -100uA, Ccbo 8pF

FMMT560, PNP, -500Vceo 1mA, -500Vebo 100uA, Ccbo 8pF

FMMT459, NPN, 450Vceo 1mA, 500Vcbo 100uA, 500Vcev, Ccbo=5pF, Vebo 7V 100uA, 625mW.

IXTY01N100 - Enhancement mode N channel, 1000V, 100mA, 200uA at 80% of Vmax - this is better than the MOSFET I used above (3).

IXTP01N100D - Depletion mode N channel, 1000V, 100mA, 250uA at 80% of Vmax - this is better than the MOSFET I used above (3).

BSP135 or BSS126  - Depletion mode N channel, 600V, >7mA, 10uA at Vmax, an avalanche diode S-D integrated so cascoding more than one stage to raise the voltage may be  possible.

Blue 1 and Green 2 probes are the separate outputs, and the Red 3 probe is Green subtracted from Blue as seen by the CRT's plates.  The capacitor C2, 3, 4 network simulates the CRT but not all the circuit's significant parasitic components. 

AL-0047-08B, SIMetrix 9.10 model.

The emitter followers impose less capacitive loading to the output than the CRT will load the circuit.  Consequently, the circuit can be made to operate at least 500KHz oscillator input shown.  With a settling time of 300nS.

The op-amp replaces the compensation network with feedback control, thereby illuminating the need for trimming the circuit in the manufacturing stage. 

C1 and C5 might be difficult to obtain, but they amount to two pieces of insulated wire twisted together.

R6 is included to relieve the conflict between outputs between C5 and C6, in this case there would not be an issue anticipated if it were not done. 

C6, R3 are compensation components that reduce the loop gain and make the circuit more stable but less accurate.

The operational-amplifier is driven to saturation, but it recovers from being driven to saturation without loss of speed, unlike much older bipolar op-amps are known to.  But it is important that the BJTs not be driven to saturation, for the same reason, which will not occur normally in these circuits, provided that the voltage input range is clipped suitably.

Some of the design steps below;

AL-0047-02A Non-inverting amp

AL-0047-03A Inverting amp


E resistors can be reduced to improve wave forms

Putting the integrated circuit idea aside and work on an all discrete semiconductor solution;

Probe 3 - the differential signal seen by the CRT's deflection plates is the significant waveform.

The circuit works best driven by a differential driver.

Model AL-0047-07b Bridge with E-followers with compensation, CRT load and differential drive.  Works >300KHz. 

This all transistor design, is about as good as the designs with integrated circuit amplifiers.  In addition, there is no extra trimming anticipated, compared to the previous circuits.  That is because feedback has been used rather than frequency compensation.

It can be seen that in this differential mode, the error on one side is corrected on the other side to produce a good differential signal, shown in green

Without the diodes D1, and 2 the pulse rating of the Base junction is not specified, and it may cause those emitter follower transistors to break. 

Without trimming at the time of, PCB assemble or the unit's commissioning input signal frequency would need to be limited.  The circuit works well without C5, and 6. See dotted trace right. The circuit above has the same outcome by having less feedback, R8 and R9.

AL-0047-01B an additional amplifier allows the circuit to work single ended input.  Its performance is the same as the one above.

A 100pF capacitor providing some differential control across R4, R10 speeds up the edges but does not improve the settling time but rather introduces distortion. 

All simulation tools retain variables that may stop the model working.  The solution is to go back to an earlier version and change it to what you wanted again.

The circuit above should be located close to the CRT connector, such as part of the connector.  The plates are driven from unsymmetrical voltages, and it would be possible to just use one driver but using two drivers in a bridge configuration would probably be best.  It turns out that the tube being World War two era design is quite insensitive and would need a differential amplifier to meet Valve or transistor rating 300V or so, consequently would also be slow because of the CRT's capacitance.  A more modern tube might be 10th the voltage and perhaps 100x faster, which is the case the best 1960s oscilloscopes were 50MHz. 

Maximum voltage may be; 500V,

Output transistor FMMT495 maximum power; 626mW @ 25'C, but from the graph given 450mW up to 60'C. 

Maximum collector current; 150mA, 500mA pulse.

R1, 2 minimum for transistor maximum power transfer; 150K = (500/2)^2 / 0.45 {(V/2)^2 / W}

Maximum supply voltage due to Vceo at max sustaining V current; 650V  = 450V + 200K * 1mA

Maximum voltage due Vcbo at maximum sustainable V current; 520V = 500V + 200K * 100uA

Time constant due to common base drive; 2uS = 5pF + 9pF [Ccbo + CRT deflection plate] * 150K. 

Time constant due to common base drive; 2uS = 2x 5pF [Ccbo] * 220K. [Emitter followers worst case]

Resistor; 1.7W = 500V^2 / 150K. 

The Emitter resistor should carry between for symmetrical speed; 

For AL-0047-11A - emitter followers; 0 to 5.3mA = 2x 400V/150K 

op-amp E drive; ~3V therefore R should 560R = 3V/5.3mA, It turns out 750R looks best.


For AL0047-08A; 0 to 4mA = 2x 450V/220K 

op-amp E drive; 750R = 3V/4mA, It turns out that 1K looks best. 

Increased transistor power consumption with emitter-follower output;  This estimate based on assumed capacitive loading.

Supply voltage reduced to 400V (calculate with; 450V)

Two transistors turned on at the same time.  5pF Ccob extra charge for the bottom transistor to dissipate joules. J = C.V^2

1uj = 5x10-12 * 450^2

At 150KHz is an extra 150mW, Therefore the circuit operate to 150KHz.

Resistor; 1W = 450V^2 / 220K. 

3BP1 CRT; x and Y plates D1 to D4 have differing capacitance to the other electrodes up to 9pF.  D1, D2 and D3, D4 have 2pF capacitance to each other.  The tube is rated at 1,500V and 2,000V.  At 2,000V, D1-D2 needs 690V for maximum deflection, D3-D4 needs 490V for maximum deflection

But having subsequently found a datasheet for the CRT, as well as being low requiring a high drive voltage to cause deflection it also has more deflection plate capacitance and this must be added to the models.  Also, the design parameters have been changed to suit the circuit options tried, they match the models well.

This circuit gave an unsymmetrical rise and fall times, but that was improved by increasing R5, 7 to 1K to give; 200KHz the input dV/dT rate needs to be limited in order to correct the distortion.  The transistors power rating is easy to work out from Thevenin's half voltage is maximum power transfer law. 

The maximum collector voltage is; 500V * 2M2/ (150K + 2M2) about 450V, which is within Vcbo limit 500V at 100uA.

Other models;

AL-0047-04A, The output resistor, has to overcome the capacitance of three transistors.

Is slower than the emitter follower option.  That option, only two transistors' capacitance has to be driven.


Simple with no compensation shows poorer frequency response.


Simple with beneficial compensation.

The Cossor oscilloscope manual pictures near the top of this page shows a number of capacitors and resistors necessary to maximise the bandwidth, to do the same as the differential amplifier in this diagram.

This looked very good but when a non-square wave input was tested in the model there turn out to be a lot of distortion;

AL-0047-12A Uses a differential amplifier.  This fast solution, but was difficult to model.

The frequency; >500KHz, +-150V Peak to Peak output.  The frequency needs to be reduced by a factor of four or more for full voltage, swing required +-300V.

The performance does not change if driven common base or common.  The circuit performance is linear, provided VOCM is set within specified voltage range.

If C5 and C6 are required, then R5 and R7 will also be required.

The transistor capacitance is the significant limitation.  Which is 2uS = 5pF * 2 x 220KR.  It will take 2uS for the voltage to rise, ready for another cycle to be driven.  This is approximately observed in all these circuits.

The feedback capacitors C1 and C7 are impractically low value, but by adding a resistive attenuator a higher value of resistor could be used.  Picture right C2, R2, R3 replace C1.

It is worth doing some sums to check the model, and that I have applied compensation in the optimal placed.  The gain is 20 = 1M / 5K the compensation capacitors are 100fF and the amplifier input capaciancwe is 2pf which is also 20:1, it seems that the compensation is applied, as is best where the error occurs.


The transistor's collector capacitance is 5pF or less (Ccbeo).  It appears that the model has used the worst case, so it is possible that all the circuits demonstrated will work better than modelled?

The low noise and screening of the oscilloscope. 

Mu-metal is best around the oscilloscope's CRT, but I did not have that, so the distance from the mains' transformers was good.  Also, the very high voltage was created with a voltage multiplier from the 250 Vac transformer, which also had a heater winding for the CRT.  The oscilloscope had mains hum in the brightness course brightness control resolved by adding a cathode resistor and no flyback suppression, which I added.  

High voltage approximately switching/sine wave power supply

AL-0048-01A CADSTAR 18.  Component values are arbitrary, so need to be calculated.

My original circuit probably had a single voltage doubler stage -1,000V, The transformer being a TV EHT/line output transformer with loose coupling and high inductance.  Possibly functions as a saturated core oscillator, with just enough drive to oscillate in a rough sine wave?

If this design were to be revised, a CCFD transformer probably could be used.  The problem with doing that in low production quantities is that you can find an excellent part cheap, but there may be a future time when it is not cheap or available any more.  Small companies just have to live with this issue.  Many but not all big manufacturers are still fairly supportive, though. 

The work on this page suits increasing the CRT voltage to 2,000V (1,800V = +450V and -1,350V, if another multiplier stage were added to the rectifier) to obtain maximum brightness before the spot blooms. 

Higher voltage low-speed drive;

There are some high voltage MOSFET types you may find a low current type, here is one; NDFPD1N150CG, On Semiconductors, N Channel, 100 mA, 1.5 kV, TO220. Note; ID=1mA at 1.2KV.  This transistor would require a lot more power than the BJT transistor solution further below. 

That increased power to over come the higher ID and this larger MOSFETs does not have a much higher reverse capacitance, so the circuit could be faster than the solutions below which use high voltage signal transistors.  If higher power consumption is acceptable, then the totem pole high voltage solution at the lower part of this page would be better replaced with a simple but differential amplifier.

AL-0011-01D Tr3 in the first circuit's capacitance Coss is <3.2pF so and the other two capacitances have trivial effect.  This circuit is likely to be faster than the BJT circuit further above, with R14 = 120K. 

These circuits are DC accurate but have poor AC performance.  If the collector load is an LED then these make accurate LED current regulators where you want to say minimise colour change due to variation in current.  In that case, I have used voltage monitoring and applied wavelength correction.  

Higher Voltage Amplifier

AL-0049-03A - With inverting op-amp, 120KHz. 

The speed-up capacitors C3, 4 are not very beneficial.

SIMetrix free version 9.10 used. Square wave input (pulse, transient options set).

There is some degree of voltage protection in that all the transistor junctions can have good but low power Zener diodes in reverse conduction.

AL-0049-01A - No op-amp, 50KHz.

AL-0049-04A - With non-inverting op-amp input, 70KHz with speed up capacitors


The all transistor solution is good, but the integrated circuit solution is better. Non-inverting op-amps one for each side might be the best solution which can be tried using LTspice, Tina TI, or the pay-for version of SiMetrix.  But these circuits do a good job any way but are improved by spending a lot of time on them. 

What I learnt and show in some of my blogs is that it is possible to mislead yourself and make a design that works on paper but depends on conditions being too precisely the similar in the final PCB assembly.  Integrated circuits provide a highly calibrated and trimmed function block, that can eliminates trimming during the manufacturing stage, and can provide the best bandwidth. 

The circuit on the right is quite good, works with single ended input to >300KHz, but is not as good as the differential amplifier IC solution further above.

This circuit is similar to the oscilloscope circuit, works well but requires very low value capacitors for integral control, which seems counterintuitive.  Cheaper than any of the solutions with ICs and is good enough.    

Engineering is not about precision but achieving precision or the desired outcome, good user feel ergonomics from imprecise manufacturing, parts and materials.  Having said that, modern electronics design is more like Lego or Meccano in that the pieces do go together without fitting (cutting and filing).

In the next page, I shall deal with compensation a little more but from a control loop perspective.  In a subsequent page, I shall restate and develop PCB to PCB and other filtering.

Your Web Browser; Right click and select open-a-new-tab to see any of the pictures above full screen.

Limitations of these circuits;

This CRT is best driven from a microcontroller DAC outputs plus digital blanking signal at lower frequency, drawing with the beam similarly to how an X-Y plotter driver works.  This CRT does not have adequate brightness to line scan a picture as a TV screen as a consequence of its comparatively low operating voltage.  It is most likely that the CRT was originally made under shared licensing agreements in about 1944 between WW2 allies.  I bought it new, but it probably was new old war surplus stock? 

The power supply circuit is accurate enough and works fine without integrated circuits, but the deflection plate amplifier is more consistent without trimming during its making by adding op-amps, so that either greater speed and lower power consumption can be achieved probably more cheaply. 

Modelling shortens the development time and with care can be used to create a good design without being lured into a false high confidence in the circuits' performance.  Take care the model does not model all the parts, even though there are models for those parts avalible. 

Discuss this see my; Blog page Electronics 

Next page; Electronics - Control Loop