DSRD pulse generator

[exnihiloest]

Drift Step Recovery Diodes seem to be the best way today to produce high power pulses (up to GW) at ns time rise, and with simple means. I have put some papers about the subject here:
http://exvacuo.free.fr/div/Sciences/Dossiers/Electronique/Pulse/

I have decided to test the concept. The principle is to "charge" the capacitance of a diode, in the normal direction, from a di/dt coming through a capacitor. When the diode is charged and the voltage is decreasing across the junction, the diode becomes rapidly non-conductive and releases very quickly its charge into the resistive load which is connected in parallel with the diode. It is matter of nanoseconds.

Here is my schematic:



To control this circuit, we need an input signal generator providing pulses of about 200ns. I use a function generator whose an input gate is validated by the output of another function generator in order to keep only one pulse on about 20. This allows for a clearer view of what occurs.
This signal controls the NPN transistor 2N3553 which can drive up to 1A at very high speed. I use it in an emitter follower configuration to drive the IRF640. It's the best way that I found otherwise it heats too much.
The diode FE2DG is a fast recovery diode whose the role is simply to eliminate the back emf from L1. L1 is made of 6 turns around a ferrite toroid core. L2 is the coupled output coil (4 turns). R3 is the load which receives the pulse from my DSRD diode FUF5404. I think it is the same as 1N5404 which is in the series 5401-5408 whose the 1N5408 is recommended for this purpose. I have plenty of fast recovery diodes, but none works well except the 5404. This one creates a pulse whose the amplitude is at least 4 times the best of my other diodes. The pulse is around 10ns (difficult to measure due to my scope limits), its amplitude goes up to 30v for a power supply of 28v. This represents a power of 19W into the load of 47 ohm, not too bad.


Input control signal (I used a x10 probe, so the voltage has to be multiplied by 10. The time scale is 1µs/div):



Signal at the terminals of the load R3 when the diode is not connected and the power supply is 20v. The view is centered on the pulse for charging the [not yet connected] diode. The pulse is narrower than the input pulse, probably due to derivative effect in the LC coupling circuit. Note that the time scale is now 50ns/div:



Same as above but now we have connected the diode: the much narrower pulse appears.



Considering that the circuit is not optimized and the IRF640 works far from its limits of voltage and current, I consider it is a success. It is very strange to observe the sharp pulse appearing when we connect the diode. Much higher power pulses can be expected by redesigning the circuit. I think it's a good tool for us.

[ion]

ex:

Thanks for the excellent work on the DSRD's. When I first learned of this effect I wanted to test it, but many things got in the way. I have some 1N5408 diodes and will build a circuit soon.

One thing to be careful of is exceeding the gate voltage on your FET when you try higher supply voltages. The gate can be driven from a regulated source (three terminal regulator) bypassed with a proper low esr cap.

The emitter follower  would normally be a safe way to drive the FET, limiting voltage to the amplitude of the signal generator, however the 100 nF cap may allow peak voltages higher than the SG voltage as it will charge when the SG is in the low state.

[Peterae]

I have waited a long time for such a circuit thankyou this is fantastic.  

Interestingly it's very similar to how i have been driving coils before, i always use an 1n4007 across my pulsed coil, i use a fet driver to drive the fet, the difference i can see is a transformer and capacitor in series with the load & diode.

One other thing i have used is UF variants of the 1N series they are ultra fast, i wonder if the UF5404 would be even better?

Thanks
ex

[exnihiloest]

ex:

Thanks for the excellent work on the DSRD's. When I first learned of this effect I wanted to test it, but many things got in the way. I have some 1N5408 diodes and will build a circuit soon.

One thing to be careful of is exceeding the gate voltage on your FET when you try higher supply voltages. The gate can be driven from a regulated source (three terminal regulator) bypassed with a proper low esr cap.

The emitter follower  would normally be a safe way to drive the FET, limiting voltage to the amplitude of the signal generator, however the 100 nF cap may allow peak voltages higher than the SG voltage as it will charge when the SG is in the low state.

Thank you for the advices which I will take into account if I go further. I built the circuit in less than an hour just to see this strange pulse and I didn't care the rules of engineering. I like the result but I'm not proud of my realization, it's horrible:    

[exnihiloest]

...
Interestingly it's very similar to how i have been driving coils before, i always use an 1n4007 across my pulsed coil, i use a fet driver to drive the fet, the difference i can see is a transformer and capacitor in series with the load & diode.

From what I read, the capacitor in series should be necessary for the pulse to occur. I have not yet asked me why.

One other thing i have used is UF variants of the 1N series they are ultra fast, i wonder if the UF5404 would be even better?
...

I don't know. It seems that the effect doesn't depend only on the recovery time. I have other diodes as fast as the FUF5404 but they give poorer results. I was in a hurry to be sure that such circuits work, so I have not studied all the matter that one can find on internet. I think we should find the answer on the net.
When I saw this kind of circuit that I was not aware despite it has been discovered 30 years ago and despite I follow the electronics news, I feel as I had moved to a parallel universe where DSRD was discovered but not yet in mine.  

[Peterae]

Great work finding this, so simple it's hard to believe 

[EMdevices]

Ex,  

thanks for sharing.   I'm wondering, have you tried different values of L2?  I'm wondering if you can obtain bigger spikes with more inductance.

From what I read, these diodes turn off quickly, ie.  resistance increases rapidly, as their forward bias charge is drained once the current reverse, and it is their speed of turn off that is of value here, because it can create sharpt di/dt which translates to sharp dv/dt with any inductance in the circuit or inside the diode itself.

what's a good application for these sharp pulses, are you building a GHz transmitter?

EM  

[exnihiloest]

...
I'm wondering, have you tried different values of L2?  I'm wondering if you can obtain bigger spikes with more inductance.

From what I read, these diodes turn off quickly, ie.  resistance increases rapidly, as their forward bias charge is drained once the current reverse, and it is their speed of turn off that is of value here, because it can create sharpt di/dt which translates to sharp dv/dt with any inductance in the circuit or inside the diode itself.

what's a good application for these sharp pulses, are you building a GHz transmitter?

EM  

With one turn more or less for L2, the pulse is not so high. It was the only point that I optimized.
The diode capacity is low and its charge is released quickly so it can provide a high pulse but in a short time. The output circuit must fit this condition by having a low impedance and matching the load resistance. This is incompatible with a high inductance.
Today I have changed C2 for 100 nF. The pulse is now 35W under 20v and can exceed 50W with higher voltage.
With 2 identical diodes in series or in parallel, there is almost no change.

The duty cycle is low. The period of repetition of pulses is very low in comparison with their duration. While the pulse width is on the order of ns, the repetition is on the order of Mhz. So for continuous Ghz transmissions, conventional radio means are better. Maybe we could load the pulse with a LC circuit resonant at hundreds of Mhz, and get a continuous signal, but we would loose the power. This circuit is intended for power pulses.
I have not yet tested the density of pulses/s that we can obtain with my setup.

[exnihiloest]

The anti-back-emf diode produced losses, so I removed it. I also realized that a supplementary coil L3 in parallel with L1 improves the process, probably due to a better impedance matching. Then I reduced the gate resistance down to 5Ω and add a coil L4 in series.  Here is the new setup:


By adjusting the input pulse width with the function generator and L3 with a ferrite rod sliding inside, the pulse rises up to 250v on my 49Ω load (I have measured it, it is an old resistance with the advantage to not be inductive but its initial value of 47Ω has a bit increased with the age).
This represents a pulse of 1,27 KW under a power supply voltage of 30v.


The circuit now:

[Peterae]

that's a slight improvement then 

[ion]

I think the circuit can be made into a simple blocking oscillator to charge the diode capacitance, thereby eliminating the need for an external signal generator.

[exnihiloest]

I think the circuit can be made into a simple blocking oscillator to charge the diode capacitance, thereby eliminating the need for an external signal generator.

Surely. The interest of the signal generator was only to have a full control of the pulse, for tests and studies purpose. But for a practical use, an autonomous circuit is preferable.

[exnihiloest]

I have some problem with the output power measurement. It could be much more. In spite that the pulses are narrow and at a low repetition rate, I observed that my resistance load heated. I suspected a problem with my probe and my scope not intended for high voltages in the ns range.

I decided to replace the load by a 47Ω resistance in series with 5.2Ω and to measure the voltage at the terminals of the 5.2Ω resistance. So the voltage being divided by 10 and the effect of the probe capacity being now negligible, I was expected for a better measurement. I measured 60v. This means 600v on a 52.2Ω load. So we would be now near 6,9 KW.

Another very interesting point is that the effect disappear when reversing L2. So it is clear that the polarity of the back emf is a key point in the power increase (in my 1st circuit, the back emf was suppressed).
Without the diode, I measure a back emf of 10v at the terminals of the 5.2Ω resistance, i.e. 100v across the load, i.e 191W. When I connect the diode, the pulse rises up to 60v, meaning 6.9KW. The load heats more but not in the ratio of the powers, likely because the diode pulse is much narrower than the back emf and consequently the mean power is not the same relative to the peak power.
I have checked the phase relation between the back emf and the diode pulse. The pulse appears just after the back emf is begining to decrease. It's not understandable because the back emf is positive at the diode cathode and should block the diode instead of charging it. Or is it because the diode was already charged? I don't understand what is going on, but the result is that there are really strong pulses that jam my radio receiver up to 200 Mhz.

[poynt99]

Be sure that you are not inducing a voltage into your scope probe.

Just something to consider.

[exnihiloest]

Be sure that you are not inducing a voltage into your scope probe.

Just something to consider.

Sure I'm not. I know that the value of the peak power seems incredible but the order of magnitude of my measurements is confirmed by the temperature of the resistance load, which heats. In my new setup, the pulses are about 10ns wide and the repetition is 20µs, this means a mean power around [peak power*10/20000] = 3.4W for 6.9 KW pulses. There is an inaccuracy due to the pulse shape which is not rectangular, but the order of magnitude is correct.

[poynt99]

OK, indeed sounds about right.

[ion]

When making very high frequency pulse measurements it is best to eliminate the ground clip lead from your scope probe and use the ground inside the spring tip connector. A very short piece of wire twisted around the ground and soldered directly to the load resistor is best.

Also use a carbon composition or other very low inductance resistor. Metal films are spiral cut to value which can add inductance.

Keep all leads as short as possible.

[Verpies]

Drift Step Recovery Diodes seem to be the best way today to produce high power pulses (up to GW) at ns time rise, and with simple means. I have put some papers about the subject here:
http://exvacuo.free.fr/div/Sciences/Dossiers/Electronique/Pulse/

The current waveform applied to the DSR diode can be generated by many different techniques, such as: a pulse transformer with saturable core, inductive flyback pulse, dual switch scheme, etc.

The DSR diode pulse generator depicted in Fig. 8 consists of two circuits with a fast power transistor Q1 (e.g.: MOSFET or IGBT, GaAs BJT, etc...) between them.
The first circuit is composed of inductive energy storage L1 and the second circuit is composed of a capacitive storage C2, inductive storage L2 and DSR Diode D1 (connected in parallel with the load resistor R1). Note that L2, C2, Q1 and D1/R1 form a series RLC circuit. Hereafter this circuit will be referred to as simply: the series RLC circuit.  
The capacitor C1 is used to stabilize the supply voltage and decrease the effective internal resistance of the power supply at high frequencies.

Before the impulse generation begins, the transistor Q1 is not conducting (the circuit is open between the drain and source terminals) and the capacitor C2 is charged up to the power supply voltage via the load resistor R1.

To begin nanopulse generation, a short pulse is applied to the gate of Q1 causing it to start conducting.
As soon as the transistor Q1 starts conducting, two events take place:
1) The capacitor C2 discharges through the forward conduction of D1 and
2) the inductances L2 & L1 start accumulating energy (the currents through them increase).

The current through L2 is inherently periodic if the series LCR circuit, formed by L2, C2, Q1RDS-ON and D1/R1, is underdamped, that is: its total resistance is less than 2(L/C)0.5. The period of this oscillation is equal to 1/( 1/LC - R2/4L2 )0.5.

After approximately one-half of this oscillation period (after interval T1), the transistor Q1 stops conducting.  See Fig.2.
During the interval T1 the D1 diode was conducting forward and charge was injected into its P-N junction.
For the manifestation of the DSR effect it is very important that the T1 interval is short enough (in the hundreds of ns) to not allow the injected charge to reach the other side of the P-N junction.

During the T1 interval, the periodic current through the series LCR circuit has reached its peak and and decreased back to zero. At the beginning of the interval T2 the current through D1 begins to reverse its direction due to the transient oscillation of the series RLC circuit, effectively causing D1 to conduct in reverse and gradually deplete the charge injected into its P-N junction during T1.

From the beginning of interval T2 the current flowing through L1 joins with current flowing through L2, D1, C1, C2 and R1. As soon as the charge injected into the P-N junction of D1 decreases to zero, the DSR diode abruptly stops conducting. This happens at the end of interval T2.

Because this abrupt interruption happens when non-zero reverse (negative) current flows through D1, L1, L2, C1, C2 and R1, a high voltage pulse appears across the D1 terminals due to the self-induction effect.
The rise time of this pulse is determined by inductance L1 and L2 and the reverse capacitance of D1.
The energy accumulated in L1 from the beginning of Q1' conduction and in L2 during the D1's reverse conduction (T2) is converted into a high electric potential appearing on the D1's reverse capacitance.

The peak power of this pulse is approximately equal to the product of the interrupted current magnitude, determined by the impedances the associated components, and the reverse capacitance of the DSR diode D1.

In order to maximize the DSR diode effect, the forward current through the diode should be lower and of longer duration, yet the reverse current should be higher and of shorter duration.
If forward and reverse current waveforms are the same then the DSR diode stops conducting when the current through the diode is equal to zero and the DSR effect disappears.
An optimal operating point for the DSR effect occurs when the DSR diode stops conducting at the peak of its reverse current.



P.S.
L1 and L2 should be air core inductors wound with a thick wire and positioned perpendicularly and away from each other.
The capacitors should be designed for RF pulse operation.  The leads of all components should be soldered as short as possible.

I am attaching a rough English translation of a Russian document, that describes the DSR Diodes and the operation of the saturable transformer driver that started this thread.

[poynt99]

Thanks verpies. Great explanation.

What's a good choice for D1?

[Verpies]

What's a good choice for D1?

The good choices for the DSR diodes are listed in the Belkin's document.
I only tried those described by Belkin and the 1N5408.

Shottky diodes do not work well as DSR Diodes, because they do not have semiconductor to semiconductor junctions, instead they have metal to semiconductor junctions and those junctions have very sharp borders.
DSR Diodes require semiconductor-semiconductor junctions (PN or PIN)

Silicon Carbide (SiC) semiconductor-semiconductor diodes (PN or PIN) could work if they were manufactured differently (more diffused), but I do not know of any that exhibit the DSR effect. There was a guy that manufactured his own experimental SiC DSR Diodes on purpose, but they were not SiC Shottky diodes - the were diffussed PN junctions AFAIR.  I can dig up his name if you want.

The ideal DSR diode structure is similar to a PIN diode that has blurred border between the P-I junction and between I-N.
In other words, for best DSR effect,  the semiconductor junctions have to be diffused (not sharp).

It so happens that the old high power silicon rectifiers were poorly made which resulted in blurred border between their semiconductor junctions and now they unintentionally work well as DSRDs.

The are very few DSR Diodes available that were tailored for this effect purposely - all of them are custom experimental devices  .

P.S.
DSR Diodes must be able to conduct very high current in reverse for apx. 20-70ns - that's the essence of the DSR effect !
The lifetime of their carriers should be long, but that is not given in the standard data sheets. In Belkin's document it is denoted by t_zh

[ion]

Verpies:

Welcome to this site. I read and enjoy all your posts on OUdotcom.

[exnihiloest]

When making very high frequency pulse measurements it is best to eliminate the ground clip lead from your scope probe and use the ground inside the spring tip connector. A very short piece of wire twisted around the ground and soldered directly to the load resistor is best.

Also use a carbon composition or other very low inductance resistor. Metal films are spiral cut to value which can add inductance.

Keep all leads as short as possible.

I have tested the method without the ground clip. There is almost no difference or a bit above. I've tried with 40v and 60v pulses at the terminal of the 4.7Ω in series with the load.

Then I changed the 47Ω load resistance with two old carbon resistances of 22Ω in series (I have not others of low value). Their real values had slightly increased with the time, representing now 52Ω included the 4.7Ω in series. I kept the 4.7Ω because it seems not to be in metal spiral. I measured without the ground clip a voltage of 53v accross the 4.7Ω resistance. This makes 586v at the terminals of my 52Ω load, i.e once again 6.6 KW.

I also replaced the 4.7Ω 0.5W with another model of 0.25W and got exactly the same measure.

Finally I put a 1N4148 diode from the 4.7Ω resistance to charge a 0.1µF capacitor with the pulses and then to measure cc. Unfortunately the presence of the diode lowers the signal for an unknown reason. I measure a cc voltage of 27v while I observed 33v pulses at the scope. Instead the pulses are 20v weaker, we see the cc measurement is in agreement with the scope measurement (and neverthess 27v cc means 298v at the load terminals, i.e. a power of 1.7KW).

It makes now almost no doubt that the power far exceeds the KW and perhaps attains 6 KW.

[Peterae]

Great stuff Ex
I had a play tonight, i don't have all the bits, i tried using a 555 timer set for the min mark/space i could get, unfortunately i could only get down to 1uS i then used a ttl fet BUK552-60A, it worked, i only had 1n5402 so put 2 in series and used a 47ohm load resistance & 6/4 turns for L1 & L2

The 1uS drive pulse was way too large, i only managed to get 30V, quiet a sharp pulse which tailed off really slowly over 250nS.

I will use a pic micro tomorrow to drive a 250nS pulse and see if i can get the ttl fet to work, i need a higher amp fet and better higher voltage diode.

[exnihiloest]

@verpies

Thank you for the detailed explanations. 
I will now try to understand why my circuit in reply#8 works well.

About the diodes, I tested two dozens of miscellaneous fast power diodes used in CRT or switching power supplies. Almost all of them work but very badly, the "pulses" being not sharp nor with a significant amplitude. Only my FUF5404 work well. I have 4 ou 5 FUF5404 and they all work in the same way. I'm surprised that there is no commercial models specially designed for this purpose.

[exnihiloest]

Great stuff Ex
I had a play tonight, i don't have all the bits, i tried using a 555 timer set for the min mark/space i could get, unfortunately i could only get down to 1uS i then used a ttl fet BUK552-60A, it worked, i only had 1n5402 so put 2 in series and used a 47ohm load resistance & 6/4 turns for L1 & L2

The 1uS drive pulse was way too large, i only managed to get 30V, quiet a sharp pulse which tailed off really slowly over 250nS.

I will use a pic micro tomorrow to drive a 250nS pulse and see if i can get the ttl fet to work, i need a higher amp fet and better higher voltage diode.

It is also important to have a very low duty cycle. Mine is now around 1/2000, otherwise the components will heat. Good luck. I will not be on line tomorrow.