Ian's software is a clever hack. It has ons drawback though, as it pushes the signal into 45 mhz filter's skirt, the signal will have sloping bandpass with more than 10db slope from one end to the other within the passband. Have I understood this correctly??

Allison, what i am seeing in my 'proto' build of driver stages only, using MPSH10's, is a flat response within 1dB BUT that's with relatively low drive levels using the Rigol at -20dBm input.

If I up the SA output level to around -2dBm (~150mVrms) which I think is the nominal level driving these stages, then the response is much worse.?? I also see severe +ve clipping levels at the base of the Q92 etc stage.
Using the sig gen (HP8640B) to drive the driver stages with 150mV, I get around 500mW out at 7MHz, rapidly dropping as the frequency rises.? Clearly, as you have said, the stages just can't deliver enough power to the FET's over the whole range.

Unfortunately I don't have any 2N3866's here but do have a few 2N4427's i could try. (500MHz Ft only though)? For the 2N2369 stage (Q90), I have the PH2369 which appears to be the same part.

Does the clipping I am seeing in the final driver stage, mean that stage needs to be re-biased with the 2N3866's?? If so what did you use?

Also, have you investigated what levels come out of the Modulator Vs frequency? It will also need to be relatively flat also for driver mods to work.

glenn
vk3pe




On Wed, May 23, 2018 at 08:35 am, ajparent1/KB1GMX wrote:

Testing note..

How I arrive at the issue of gain vs frequency.

One was on paper using the well know EMRD feedback amp spreadsheet.
Every try said not possible unless the device has a FT of greater than 600mhz
THe simple test is we need a gain of 39 (~16db at 3mhz) and at 30mhz we need?
a device Ft of not less than 1170mhz based on the Ft/F rule of thumb.? We can
however get away with a bit less with only a small decrease in gain.

The other was a mod to the board and measure:
I removed the IRF510s and added a trans former to go from the gate connection
to 50ohms (1:1), added 10db 1W pad for safety (protect the Rigol). and to keep the?
mosfet out of the test.

Lifted C1 to isolate the TX amp from the LF and hooked a piece of coax with a.1uf cap at the and to
the input of Q90.??
RV1 set to 100% so we see total stage gain.
Enabled TX mode with out Raduino, jumper T/R to 12V..
Set up RIgol DSA 815Twith tracking generator for -20dbm output from 3 to 30 mhz.
I didn't do a screen capture though I should have.
( I used lower drive as I'd seen gain compression in driver and pre-driver at higher
output levels at higher frequencies. this is device HFE related at higher currents.
this makes the system look worse.)

The result was about 45DB gain at 3mhz falling to 34db at 30mhz. A drop of nearly 10db.
The slope of the curve was such that at about 7mhz we were already dropping by 2db.
Power out of the driver for that setup was .255 Watt at 3mhz falling to .023W at 30.
We could get more power by cranking up the trancking generator output at lower
frequencies but gain compression at higher power levels limited us at?14mhz and
above.

Adding emitter caps and inductors to the feedback helped but the result was still
well short.? The net effect is the curve was flatter to about 13mhz then resumed?
sloping down.

Why is this.? The 2n3904 has a FT of 300 (nominal) and the rule of? thumb is ft/f=Beta
BEta is the maximum attainable gain for the stage.? So at 3mhz the result is 100, lots
of gain and its tamped down with feedback to 39 (~16db).? ? At 30mhz the result is 10!
10 is 10DB of gain...? ?Paralleling transistors is supposed to help that but the result is
typically a 3db improvement at higher frequencies. So for 4 stages at approximately
16db per stage at 3mhz showed?48db (Not allowing for transformer losses) but close
to the result of 45 DB gain.? ?However at 3mhz we are supposed to get 30db and
did get 33 so we are doing better than predicted?but well down from 3mhz.

Adding the various caps and inductors in the feedback to the q90, pre-driver, driver made
the curve flatter but at 14mhz we are now 3db down and would end up at 35db down at?
30mhz.? Clearly the 2n3904s were not cutting it.? Its hard to get more than? a beta of 20
out of two in parallel and we need an effective beta of at least 39.

This is why higher frequency transistors that work at higher currents are being pursued.
Devices with FT greater than 650 mhz gets use within about 3-4db without circuit tricks.
3Db flatness means at 3mhz say 10W and 30mhz 5W.? Adding tapered feedback helps
some so we should see near 7-9W at 30 and so far my results confirm that using
2n2369 for Q90, and the pre-driver one 2n3866,? and driver two 2n3866.? ?All of those
have an Ft of 600mhz or so and power handling suitable for the stage they are in.
The problem is 2n3866 and 2n5109 are not cheap.

NOTE: if as was, we get about 4W (at 30mhz) from the IRF510s with the measured drive
it confirms the MOSFET is being starved!.

Putting the irf510s back in and grabbing the 50W 30db power attenuator agrees
nicely as the IRF510s at 3mhz were netting about 16db of gain and at 30mhz a
respectable 14db of gain.?

Conclusion is we need uniform gain and the ability to deliver more than .5W
(about .7 would provide good headroom).? ?The IRF510 has enough gain.

Now I wait for lower cost parts.


Allison

Allison,

Getting the analog stuff to where it is flat from 3.5-30mhz is preferred if the mods are easy,
we much appreciate your efforts in this in this regard.

However, gain per band can be done with zero additional hardware,
a rare case of being very very cheap:
? ??/g/BITX20/message/50217
? ??/g/BITX20/message/50184
The KD8CEC software already has this feature fully implemented.

I am curious if you have ruled this method out for some reason.
You'd think it might mangle the audio frequency response some,
but I can't hear anything other than reduced amplitude as I take it down the skirt.
That's a fairly broad filter.

Jerry, KE7ER



On Thu, May 24, 2018 at 05:25 am, ajparent1/KB1GMX wrote:

Gain per band has been widely used and work save for its rarely cheap and tends
to add a lot of circuitry.