I had trouble with the GDT signal(signal ripples) for half a year.
Finally, I found the reason is that the UCC2732x driver chip (China’s product) has 3 pin does not work as enable pin.(no function pin)
The way to fix this problem is: use this schematic below

ChipTC4422 can work up to 18VCC so they provide more power to drive the GDT.

pipple GDT wave

Normal GDT wave

Accessory:
-A coil can work OK on primary feedback mode
-An oscilloscope
Schedule of doing:
1-   winding 1 turn of wire around the wire that primary current going through,
hook the probe of the oscilloscope on 2 terminal of that wire
image 1
2-   take the L2/C2 out of their place
3-   let the coil running on normal DC bus and  short burst duration(20us) primary current can reached  500Apk or more depend DC bus and burst duration.
4-   observe or take a picture the primary current wave ( at 1v/div setting on oscilloscope)
image 2
In this case the result is 9.5*5/8=5.9375us f1=168.42kHz
That is an safety measurement because the probe of oscilloscope not connect to the circuit 

Content in this topic overrides some of my old topics in the form of an update action
      Power output/primary current according ratio f2/f1preset

This table construct by run simulating 4 models of Tesla coil, remember that in 6 columns it always running in-tuned (frequency in L1=frequency inL2=ft) I have verifier once by my test coil in practice , It run immediately and pretty good at column 1 . All of 4 models give best performance at f2/f1=1(preset)
*In column1;4;6 first number is power output peak max at the end of burst duration, 2nd number is primary peak max Apk at according position.
While DRSSTC  running, it always in-tuned regardless of :
-   split frequency factor
-   Situation of  load “no load or full load”
Therefore, the meaning of the tuning is  preset f2 higher than F1 15% then it will be into tune when the sparks is stronger no longer trendy, now is the job of finding out which value of frequency split up factor should be preset, will give the best performance. It is best include the highest power output according to the power input ratio. The results are simple:
 “To get the best tuning, let refine f2 = f1”
For classic Tesla coil it is different

Note: Row3 first  number Kw peak max.  Second  number KA peak max
This table proved  that  why need to preset f2/f1=√1.3=1.148. Because it give best performance in this case.
And it also explain why during hundreds of years nobody has question about that.
May be at other of L1;L2 value this number change a little or not?

The Tesla coil is as a function that has many variables, so it gives a lot of results, countless results. This problem makes it very easy to achieve average results but to get perfect results is very difficult.
This table helps to show great prospects for success, from these points it is hoped that just fine adjustment is enough.
Call:
F1=1/2π√L1C1 primary resonance frequency (static)
F2=1/2π√L2C2 static secondary frequency (static)
Ft=Tesla coil resonance frequency measure by oscillosope+antenna (dynamic)
F1 and F2 number by calculating or measuring while only one coil run seperately
√(L1.C1/L2.C2)=F2/F1 [/i](amount of split  frequency from F2 against F1) temporarily call slit factor from 1 to 2
Quick access good tuning point  table (DRSSTC, L2/L1=10000)

What is this table built by?
* F1 and F2 are parameters of each individual coil.
* When Tesla coil is physical formed (in coupled) F1 and F2 no longer exist, Ft replaces them, both coils run only on one frequency Ft regardless of light or heavy sparks.
* Run simulating the schematic and increase split  factor  from 1 to 2, the AC sweep curve to find the resonant frequency of the Tesla coil is the shape of the horse saddle including the upper and lower poles then gradually taking the nose shape that has only one resonant frequency point, the output power of the Tesla coil increases proportionally and then changes in the downward direction, at the highest point of this curve is a good tuning point. (points with the highest power output)
     Compare with common theoretical:
Quote from http://weber.itn.liu.se/~stegu/onetesla/onetesla_tuning_guide_draft.pdf
                                                    Getting into tune
 “Measure your resonance frequencies on minimum power, or on an unpowered coil using a signal generator and an oscilloscope. If you are lucky, your primary frequency is about 15% lower than your secondary frequency. In that case, you're basically fine. Your coil does not need any strong tuning adjustments. The reason for having different frequencies for the primary and secondary is that the secondary frequency will shift down somewhat during operation when there is a long spark present. That spark is a conductor that will add to the capacitance of the top load and reduce the resonance frequency by about 15%. See the section “Deliberate detuning” above.”
This way of speaking is present everywhere in the theoretical resources of Tesla coils 
In fact its nature is so, it is necessary to spilt up them first with an available split factor and then they will be unique after combining. Completely F1 and F2 are not unique by strong sparks.   The action of this tutorial is OK, but explains why to do it, it's not right.
Image 1: beginning of the spark from 13us to 23us the periods is 6.74us

Image 2: at the end of spark from 1130us to 140us the periods is 6.94us

*The frequency at the end of spark lower than frequency at the beginning of spark approx.  3%

Image 3: sine wave in primary and secondary are  the same 20 cycles in 140us, this image showF1 and F2 no longer exist, Ft replaces them.
Note:
*Number L2/L1=10000 is reasonable in terms of physical and final result. If it is become 5000 the output voltage and power output drops to 66% lower than before in almost the case. Row 1 and row 2 this number are 7500 and 6250bucause L1 cannot <4uH.
*For classic Tesla coil the number of column 3 maybe change a little.
* In the low-frequency area <80kHz, I have to change the load resistance from 5p to 10p or higher to get high power, there is no other way to do it. I think that sparks create their own power by increasing its capacitance according to its length or anything else that I don't know exactly.
* High inductance coils need higher power to run. So global use and higher DC bus.( Row 11, row12)
Thank you all guys.
Sicerely.

    After run simulating several hundred  times with old schematic , I am very tired because after chang the value of  L,C i have to find  it’s resonance frequency at that case then set the periods of input signal  matching the loop of oscillations, that is to do many things.
     I am very happy when succeed in design a schematic that it can run oscillation by itself every change any value of component. This matter is very helpful because reduce 80% thing to do in learning Tesla coil.
 I think this schematic also can apply in practice ,but need to modifier something at GDT transformer, it is simpler the electric circuit a little, especially haft bridge running  in single bus, make more simple and advantage in rectifier and filter.
    Now, I think the big matter need to make clear is find out the best equivalent circuit of the corona, the spark in lite, medium and heavy load. This thing helpful in making the simulating give the result more accuracy.
schematic
power in/power out
power in L1/power in L2
IBGTs's current
Top load voltage

Very interesting! Thanks for sharing  I'll try running it against the simulation. I think i will study a lot from this topic.
Thank you!

This simulation image illustrates almost information of energy movement in a spark gape Tesla coil.
Image 1


The red region is energy charged into capacitor C1.
The orange region is energy discharge fromC1 to L1, it is a supper high surge power but it’s duration only 1us.It’s power is 1.15Mw peak but it’s energy is only
P=1/2 V*V*C1=1/2*5000*5000*80/1000000000=1joule
 In this case it run on 2nd harmonic frequency. I don’t know why software Orcad 17.2 lite run simulation like that.
In the no color fill region in the orange sine wave and purple sine wave. These halt cycles 1;3;5 are energy fromL1 going back to C1.These halt cycles 2;4 . .are energy fromC1 running to L1.
Image 2

Reactance power in L1red wave and Reactance power in L2purple wave.
Image 3

Reactance power in L2and dissipation power in the spark.
 In these waves the current and the voltage shifting  90degree,this is reactance power also called no energy power(ccosɸ =0) almost energy are saved and a tiny loss in dissipation and some in RF radiation. This action generate a strong electromagnetic field around the coils these thing is the agent for the transfer of energy through space or at least from primary to secondary coil. A super high power reciprocating movement back and forth bettwenL1;C1 and L2;C2.
The yellow region shows power dissipation in the spark released from the toroid.
Action charge into C1 and discharge fromC1 can say no loss energy. But by the transfer action maybe more energy loss so the yellow region seem smaller than orange region a lot.
This image also shows that these waves are complex by reason they are not the same phase.
One should emphasize that in this original Tesla coil the current in the primary and secondary coil shifting 90 degree approx. That is a difference between it and DRSSTC.

I focused on design electrical circuit that could create the burst pulses length look like ideal square wave as figure1 (Marty’s image, no ripple wave).

I spend a lot of time to exclude the ripples of GDT signals after try, try, but the result was a disastrous failure, but somehow managed to catch it a part.
What is a disastrous failure? Fist,I try to make a circuit can generate square wave symmetrical and try to coupling direct to GDT but cannot and several driver chip UCC37321 fail.
I am exasperated, so I run simulation to find out the reason.(image2)

The result of simulation shows that voltage wave look very good

but current is deviation to one side of polarities (image3) and seriously in the long duration signals (image4)


Then I must be used capacitor 1uf for DC cut (image 5) switchU1 start burst lengthU3 stop burst length, U2 switch simulate UCC37321 pull to GND. These switches simulate driver chips working in interrupt mode.
Right after the capacitor the ripples appear by 2 parts: front ripple and rear ripple they have the periods 160us . They created by capacitor 1.5uf and inductance 440uH.in serial circuit. (image5) reduce ripple circuit coupled.(parameter from my real circuit with the coil 5turns on ferrit core 5mmx12mm cross section)

Solution to reduce part 1 of ripple is rather simple the amplitude limiter used zener diode can gives good effect to do that (image6) reduce ripple circuit coupled, front ripple was almost excluded

Part2 of the ripple I not yet find out the way to reduce it. In practice it do not causes the bad effect to run Tesla coil.

Hi  T3sl4co1l,
Thank you for your reply
I used 6.5mmx13.5mm cross section core and winding 26 turns of wire may be higher double of turns than before,the result of the test look like no change as these images

 May I ask you upload your GDT signals at output side? I don’t understand why the wave form of Marty has not any ripple at the front and the rear of the burst pulses length?

Hi Phoenix,
I apply Steve’s schematic for my circuit in image 5

C5 and reactance of GDT cause this ripples include input capacitance of IBGT but
I don’t know the gate signals that Marty reply me, why it is so beautiful like that?
quote (Marty)
Here is output signal (which is going into gate of 1 IGBT):

Thank you!