Topics

Amplifier chain design - best practices


Jeff Omundson
 

For me at least, this is more conceptual, but I am interested in understanding more after reading about the issues  with the uBitx power amp driver and predriver problems.  What are the best practices for designing this portion of a transmitter when it is to cover such a wide range of frequencies?  I imagine it also applies to a monoband radio.

This is what comes to mind based on my limited knowlege and what I've gathered from reading stuff here.

1)Keep amplifier intos and outofs away from each other.
2)Keep amplier gain stages reasonable.  Add more stages instead of upping gain.
3)Pick the right device for the desired frequencies.  Obviously 2n3904's don't work for this application.  Does picking a part with an fT about 40 times the highest desired frequency a safe bet?  Can you pick an fT that is too high (like a npn transistor with and fT of 10 gHz for a 3-30mHz radio)?  I ask this last bit after looking through the digikey transistor offerings...

I suppose any device needs to be tested and tweaked for its application.

4)Keep rf wires/traces away from each other and from the dc path.  

Makes me wonder how the ft817 does hf, vhf and uhf...  I suppose it could be a good exercise to look at how others have done it successfully, like elekraft with the k2.  It will be interesting to see how the QSX does it.

Would making a transceiver optimized for one band and then making transverters for the other bands avoid some of these problems?  I'm thinking of what is outlined in the From Crystal Sets to Sideband book.

-Jeff
AG7NW


jim
 

Just as a suggestion, before you start ANY mods ...Consider replacing RV1 with a 10k or so pot .... All the preamp gain is getting eaten up by C82/100 ohm RV1 to ground .,,

Sometimes the simplest things are hardest to see

Jim




On Friday, September 21, 2018, 9:19:44 AM PDT, Jeff Omundson <jeffomundson@...> wrote:


For me at least, this is more conceptual, but I am interested in understanding more after reading about the issues  with the uBitx power amp driver and predriver problems.  What are the best practices for designing this portion of a transmitter when it is to cover such a wide range of frequencies?  I imagine it also applies to a monoband radio.

This is what comes to mind based on my limited knowlege and what I've gathered from reading stuff here.

1)Keep amplifier intos and outofs away from each other.
2)Keep amplier gain stages reasonable.  Add more stages instead of upping gain.
3)Pick the right device for the desired frequencies.  Obviously 2n3904's don't work for this application.  Does picking a part with an fT about 40 times the highest desired frequency a safe bet?  Can you pick an fT that is too high (like a npn transistor with and fT of 10 gHz for a 3-30mHz radio)?  I ask this last bit after looking through the digikey transistor offerings...

I suppose any device needs to be tested and tweaked for its application.

4)Keep rf wires/traces away from each other and from the dc path.  

Makes me wonder how the ft817 does hf, vhf and uhf...  I suppose it could be a good exercise to look at how others have done it successfully, like elekraft with the k2.  It will be interesting to see how the QSX does it.

Would making a transceiver optimized for one band and then making transverters for the other bands avoid some of these problems?  I'm thinking of what is outlined in the From Crystal Sets to Sideband book.

-Jeff
AG7NW


Curt
 

Jeff

great question on what's a good transmit path design.  since I have the K2 manual here --

predriver is a 2n5109, feeding a (;3 transformer on a FT37-43

driver is a 2SC2166 NPN feeding a 12:8 transfomer on a FT37-43

maybe some soft equalization network here ?  that's all I see that might seek to flatten RF performance

the PA is push-pull pair of 2SC2166 into a transformer to achieve the desired output power to feed 50 ohms

these devices fit the original K2 price target but not now the primary, global uBITX users.

the FT817 uses different PA devices for VHF/UHF than HF I think.  I imagine the VHF/UHF PA module may cost more than a uBITX.  obviously the climb for HF to 50 MHz requires more savvy PA devices. 

on the uBITX PA, when details are revealed on the IRF510 PA designed by Allison and Hans for the QSX there will be something of value here for v5, and to see if we can improve use of the v4 board. 

73 Curt


ajparent1/KB1GMX <kb1gmx@...>
 

On Fri, Sep 21, 2018 at 09:19 AM, Jeff Omundson wrote:
1)Keep amplifier intos and outofs away from each other.

This also applies to all RF systems.  

2)Keep amplifier gain stages reasonable.  Add more stages instead of upping gain.

This is a serious and common error, forcing or wishing for gain that is not predictably there.
Conservative design rules insure that.

3)Pick the right device for the desired frequencies.  Obviously 2n3904's don't work for this application. 

First the 2n3904 can if one does not apply wishful thinking and employs it within the devices capability.

***Does picking a part with an fT about 40 times the highest desired frequency a safe bet? 
Oddly the answer is yes if the stages has to deliver at least part of the gain desired.  For a 
30mhz amp with a 20db stage gain you will need an Ft of something approaching 3000mhz!
The real question is does it makes sense when two 3904s at 10DB each can also do that 
at 30mhz.  

***Can you pick an FT that is too high (like a npn transistor with and fT of 10 gHz for a 3-30mHz radio)?  
The answer is yes and maybe.  IF you using resistive feedback amps (nearly the entire transmitter and both IFs)
the required FT is a product of the gain and the frequency so at some point you really have to ask is that much
gain per stage reasonable and stable.  IT might be prudent to back down the gain and use a more reasonable
transistor.  Also for higher power stages the cost for a supergain high power device might break the bank.

A lot of QRP designs more so older one try to be minimal and get the most out of a very few parts.
If you making one this is ok as you can dial it in and hand pick parts.  For quantity builds this is terrible.
You want every one to be the same as the last.  So care in not trying to get 101% of capability is advised..
This shows more often in transmitter circuits as out of a group most doe the expected watts, a few do
more and a bunch never do that.  Multiband transmitters due to bandwidth, power and layout require
greater attention.

4)Keep RF wires/traces away from each other and from the dc path.
Again this works for RX and TX circuits.  We do not want o induce unwanted signals into the receiver.

From Crystal Sets to Sideband, a very good starting place.  My Favorite are Solid State Design For Radio Amateurs,
and Experimental Methods in RF Design (Both ARRL).  Neither are light bits and do require several reading through
passes to pick up that what was mentioned here is often elaborated and related to things there.

***Makes me wonder how the ft817 does hf, vhf and uhf... 

Lots of tuned filters, careful layout the use of the chassis as cooling, shielding, and mechanical support.

Its worth looking at Elecraft K1 and K2 as the manuals and detail in them is very good as examples.
I'd add KNQ7A  and the KD1JV Sopbucket 20 as a simple monoband, and a few other as well.  Generally
the designers are creative people so watch for solutions to problems not outwardly discussed as its
often not featured but important.  Cooling and layout are often not obvious without analysis.  Same can
be said for component choices.


Allison


ajparent1/KB1GMX <kb1gmx@...>
 

Jim,

Replacing RV1 with a 10K is not good as it would put all the adjustment in the last 5 degrees of rotation.
A 500 ohm may be better.

But you begging 1 db of gain increase for what reason?   Replace the 3904 in that stage prior and the
performance jumps a greater degree and fixes the actual problem there.

Allison


ajparent1/KB1GMX <kb1gmx@...>
 

Curt,

Those would be good solutions if you can get 2sc2166s at acceptable price and quantity
without incurring the problem of fakes.  Its part of why the K1 was discontinued.

We cannot rely on unobtainium.   Not that that was not a good solution but the challange
is to find widely available inexpensive parts.  As a professional designer and home builder 
I have an acute realization for that.

HINT:  3904 can be used it employed at a performance level it can do.  The other is
board layout in a multiband radio is more critical.  Particularly the PA area.

Allison
Engineers build what they want with things they can get.


jim
 

 Spice it ...I did (using QUCS) .... Nearly a direct short to ground off the high impedance point of the amp (before the change) ...and you still have a linear control of the following amp chain (just checked it on my bench ...100 mV p/p in at head of LPF where it comes from the mixer to top of RV1 (3.0 V p-p) ...~30 db gain on modified Q90 circuit

Jim


On Friday, September 21, 2018, 10:25:26 AM PDT, ajparent1/KB1GMX <kb1gmx@...> wrote:


Jim,

Replacing RV1 with a 10K is not good as it would put all the adjustment in the last 5 degrees of rotation.
A 500 ohm may be better.

But you begging 1 db of gain increase for what reason?   Replace the 3904 in that stage prior and the
performance jumps a greater degree and fixes the actual problem there.

Allison


jim
 

I also too have the K2 manual for my unit ...Looking a "little further in front" you will notice 5 ea. band-pass filters ...Not dc to daylight transmit path ... if you want K2 performance, buy a K2 @ $800 to $1200 ...(glad I got mine when they were cheap)
elsewise work with whacha got

Jim



On Friday, September 21, 2018, 10:15:31 AM PDT, Curt via Groups.Io <wb8yyy@...> wrote:


Jeff

great question on what's a good transmit path design.  since I have the K2 manual here --

predriver is a 2n5109, feeding a (;3 transformer on a FT37-43

driver is a 2SC2166 NPN feeding a 12:8 transfomer on a FT37-43

maybe some soft equalization network here ?  that's all I see that might seek to flatten RF performance

the PA is push-pull pair of 2SC2166 into a transformer to achieve the desired output power to feed 50 ohms

these devices fit the original K2 price target but not now the primary, global uBITX users.

the FT817 uses different PA devices for VHF/UHF than HF I think.  I imagine the VHF/UHF PA module may cost more than a uBITX.  obviously the climb for HF to 50 MHz requires more savvy PA devices. 

on the uBITX PA, when details are revealed on the IRF510 PA designed by Allison and Hans for the QSX there will be something of value here for v5, and to see if we can improve use of the v4 board. 

73 Curt


Curt
 

I was only citing K2 as example, yes now unobtainable or costly parts as I alluded to. Some of its circuits, like the uBITX, have legacy to discussion in those 2 W7ZOI et al books. The older one has lots of interesting concepts, but many unobtainium parts. What we can get cheap enough can change rather quickly.

I respect many folk around the world cannot afford a K2, hence the uBITX paradigm. Lets see how that new QSX related IRF510 PA works when it comes out.

73 curt


Jeff Omundson
 

Thank you for the thoughtful responses!  Great ideas to digest.  I have the Solid State Design book but have not worked my way through it.  May be that is homework :-)  The explanation at the start for biasing transistors always made more sense to me than what I read in other books, including EMRD.  It is a little hard that parts become obsolete and so the projects might be difficult to build as is.  I suppose it could be broken down into smaller bits and then modelled in SPICE with best guesses at replacements for hard to find bits.

I thought buying the uBitx would be kind of a simple and unchallenging project radio.  I bet a lot of people didn't expect it to take us where it has!I haven't done any of the work to make it better, but I appreciate all the work that is being done.  It has prompted me to think deeper about multi band transceivers and it has been interesting to observe the community (and certain individuals in particular) work out solutions.  I think we are all learning more than if we were content to just play radio on the major commercial offerings.

Thank you to Ashar and all of you who have spent so much time working on this and sharing the knowledge!

-Jeff
AG7NW


jim
 

thats my take on the uBitx ...Simple ie cheap parts ..Easy to work on ..Its all right on top ..spread out with BIG surface mount parts ... Play to your hearts content

Jim
 

On Friday, September 21, 2018, 12:21:03 PM PDT, Curt via Groups.Io <wb8yyy@...> wrote:


I was only citing K2 as example, yes now unobtainable or costly parts as I alluded to. Some of its circuits, like the uBITX, have legacy to discussion in those 2 W7ZOI et al books. The older one has lots of interesting concepts, but many unobtainium parts. What we can get cheap enough can change rather quickly.

I respect many folk around the world cannot afford a K2, hence the uBITX paradigm. Lets see how that new QSX related IRF510 PA works when it comes out.

73 curt


ajparent1/KB1GMX <kb1gmx@...>
 

On Fri, Sep 21, 2018 at 02:58 PM, Jeff Omundson wrote:
It is a little hard that parts become obsolete and so the projects might be difficult to build as is. 

I have booth books and more even old copies of HamRadio, 73, CQ, Pop'tronics.  Its rare I
find a circuit that can't be built because the device usually a transistor or tube is unobtainium.
As a experimenter that's part of the game to figure out what part that is available might be
a good match.  For the most parts Diodes are easy, most transistors are not magic either.
I still pull down SSD and play with ideas.  Spice, rarely and I've driven that and many others,
then again I can drive sim or a slide rule its just numbers and if you don't understand what
their significance its data not information.

The rest of the parts are coils, caps, and resistors and they are fare from rare.  There may
be exceptions but they are far from the rule.  

True thing that ubitx would take people into parts of tech they would never have gone.
Playing with the user interface (raduino firmware) has been the big fun for me.  I was
shocked how few understand superhet radio being a near 90 years old and well
understood.  I suspect that from it being new to people that have never worked with it.
So they get to learn it and all the little things that go with it.

Allison


Howard Fidel
 

I've been sitting here on the sidelines for a while. I was hoping to have a complete solution finished by now. However, life gets in the way. I unexpectedly got 2 consulting jobs, plus my teaching job has eaten up most of my free time. However, I think it is worth putting  my 2 cents in now.

I will address the harmonics issue. We need to look at the root cause of the problem. The harmonics occurs in CW because it is direct conversion (or no conversion?) CLK 2 runs at the operating frequency. This clock comes from the infamous SI5351 chip that is also causing the spurs. Well, this clock is a square wave. The mixer it drives is unbalanced in CW and the clock just passes through and goes through a lowpass filter. (L1-L4 etc.) This removes frequencies above the 10 meter band. As most of you know, a square wave is made up of only odd harmonics, so we have the 1st, 3rd, 5 th etc. harmonics coming out. So for the 80 meter band operation many harmonics will pass through this lowpass filter get amplified and need to be filtered out by the output lowpass filter. We see fewer harmonics from 40 meters, mainly the 3rd since the others are removed by the first lowpass filter. At 20 meters and above the signal coming out of this filter looks like a sine wave, it has all the harmonics removed, hence there are no harmonics coming out of the transmitter. Garbage in garbage out. There are many ways to solve this problem. To date most everyone has blamed the poor output filters. We could also fix this problem by insuring we always drive the transmitter section with a sine wave to begin with.

So, we could modify the L1-L4 so that its cuttoff frequency is about 8 MHz for use on the 80 and 40 meter bands and stays what it is for the other bands. This will remove the harmonics earlier in the chain.

 Another, easier  way is to heterodyne in CW as well and go through the 45 MHz filter. So we need to move the unbalance circuit from the above mentioned mixer to the CLK1 mixer. Drive this mixer at 45 MHz and clk2 would would be the same frequency as we use for SSB and the harmonics would be gone. To do this all we need to do is lift R105 and so it stands up on the C1 side pad and place a jumper wire from the open end of the resistor to pin 6 of T4. Of course the software needs to be updated, but now without any filter changes, and with a very simple hardware change the harmonics are history!

Howard


Jerry Gaffke
 

Howard,

Yes, there are problems up front around the first and second IF's that create harmonics and spurs.
Yes, replacing the 30mhz LPF with band specific filters could help with both spurs and
harmonics during transmit, though does complicate what is trying to be a simple rig.

Here's my take on a couple of your points.
Unfortunately, I have not found time to do any experimenting on this stuff myself.

> To date most everyone has blamed the poor output filters. We could also fix this problem
> by insuring we always drive the transmitter section with a sine wave to begin with. 

The output LPF's are fine, it's just board layout and how the relays got wired up
that allows way too much blow-by.  When the board is hacked to clean this up,
transmissions are fine with respect to harmonics.  The power amp will generate
some harmonics (and IMD) on its own, especially when pushing for more power on the
upper bands.  Many CW rigs operate class C (so a square wave at the final),
simple output LPF's work fine even then.
 
> This clock comes from the infamous SI5351 chip that is also causing the spurs.

The primary cause of spurs seems to be in the diode ring mixers, 
too much signal and not enough local oscillator.
The fix is less IF gain and higher local oscillator levels into the mixers.

The Si5351 is having a hard time dealing with the 50 ohm 6dB pads it's trying to drive,
but otherwise doing pretty good.  Crosstalk due to the heavy loading is causing
some carrier leakage and perhaps a few extra spurs.  
Would be interesting to know if anyone's tried this:  https://groups.io/g/BITX20/message/35206
discussed further here:  https://groups.io/g/BITX20/message/59220
Unfortunately, I have not found the time (or the logic analyzer) to evaluate it.

Do dig in if you can and let us know if you find some new tricks to help clean the rig up.

Jerry, KE7ER


On Fri, Sep 21, 2018 at 07:35 PM, Howard Fidel wrote:
I've been sitting here on the sidelines for a while. I was hoping to have a complete solution finished by now. However, life gets in the way. I unexpectedly got 2 consulting jobs, plus my teaching job has eaten up most of my free time. However, I think it is worth putting  my 2 cents in now.

I will address the harmonics issue. We need to look at the root cause of the problem. The harmonics occurs in CW because it is direct conversion (or no conversion?) CLK 2 runs at the operating frequency. This clock comes from the infamous SI5351 chip that is also causing the spurs. Well, this clock is a square wave. The mixer it drives is unbalanced in CW and the clock just passes through and goes through a lowpass filter. (L1-L4 etc.) This removes frequencies above the 10 meter band. As most of you know, a square wave is made up of only odd harmonics, so we have the 1st, 3rd, 5 th etc. harmonics coming out. So for the 80 meter band operation many harmonics will pass through this lowpass filter get amplified and need to be filtered out by the output lowpass filter. We see fewer harmonics from 40 meters, mainly the 3rd since the others are removed by the first lowpass filter. At 20 meters and above the signal coming out of this filter looks like a sine wave, it has all the harmonics removed, hence there are no harmonics coming out of the transmitter. Garbage in garbage out. There are many ways to solve this problem. To date most everyone has blamed the poor output filters. We could also fix this problem by insuring we always drive the transmitter section with a sine wave to begin with.

So, we could modify the L1-L4 so that its cuttoff frequency is about 8 MHz for use on the 80 and 40 meter bands and stays what it is for the other bands. This will remove the harmonics earlier in the chain.

 Another, easier  way is to heterodyne in CW as well and go through the 45 MHz filter. So we need to move the unbalance circuit from the above mentioned mixer to the CLK1 mixer. Drive this mixer at 45 MHz and clk2 would would be the same frequency as we use for SSB and the harmonics would be gone. To do this all we need to do is lift R105 and so it stands up on the C1 side pad and place a jumper wire from the open end of the resistor to pin 6 of T4. Of course the software needs to be updated, but now without any filter changes, and with a very simple hardware change the harmonics are history!

Howard


Tom, wb6b
 

When I finish with my homemade SA, I made a list of a few things I'm going to try.

1). Improve ground layout. I plan on taking a Dremel tool and cut a line across the ground copper flow between the PA side and the IF amplifier side. Then I'll find a better common ground point for the two sides. This may reduce the ground loop from the PA feeding into the IF stages.

1a). Experiment to see if using an additional copper clad board, mounted under the uBITX, as a common ground island with various critical points of the circuitry grounded with short straps to the ground island helps at all.

2). Reduce the audio drive level to about where the uBITX currently puts out 1 watt on the lower bands. Then add an additional pre-driver stage before the PA to bring the power back up to around 10 watts. This may make life easier on the IF stages and the mixers.

3). Play more with moving the DC power feed points around. 

4). Rip up the filter relays (Allison's suggested filter fix) and rewire or just go with an external filter.

5). If I go with an external relay switched filler, I may buy a 10W amplifier kit, with a pre-driver, that only requires around 100 mW of drive, and tap into the TX drive chain ahead of the PA on the uBITX board. I'll do this at a point that provides enough gain so I can get 10 watts out with reduced audio drive. I might put both the amplifier and the relay switched filters in the same box. May use the Kees relay board with QRP Labs filters. 

6+) Whatever solutions that have been suggested on this forum that help solve the problems not address by the above list. 

I'm willing to accept the current uBITX may not work on every band, and have other limitations, so I my not do every mod that comes along, if it only improves a band or mode I'm not interested in. 

When a new uBITX revision comes out, I'll likely buy one. As the cost of two uBITX transceivers averaged together (one all bands and one a few bands) is still a pretty good price for two radios I the shack. I do hope any new uBITX will have the same form factor, so it can be dropped into any case people have already built for their current uBITX. 

Tom, wb6b


jim
 

This ...I measured 0.6 volts P-P on the junction or R102/R103 ...thats about 0 dbm ...Not enough for a 'level  7 ' mixer  (more nearly a sine wave than a square wave  or kinda a sineish square wave)

Jim

On Friday, September 21, 2018, 9:21:53 PM PDT, Jerry Gaffke via Groups.Io <jgaffke@...> wrote:


Howard,


 

The Si5351 is having a hard time dealing with the 50 ohm 6dB pads it's trying to drive,
but otherwise doing pretty good.
Jerry, KE7ER




Michael Maiorana
 

Jim,
Your post made me curious. I have an early K2 that I bought and built in 1999. Looking at my invoice, and adjusting for inflation, the K2 is about $100 cheaper today (100 2018 US dollars). 
Inflation is a thief. 
Mike M.
KU4QO

On Fri, Sep 21, 2018 at 2:55 PM, jim via Groups.Io <ab7vf@...> wrote:
I also too have the K2 manual for my unit ...Looking a "little further in front" you will notice 5 ea. band-pass filters ...Not dc to daylight transmit path ... if you want K2 performance, buy a K2 @ $800 to $1200 ...(glad I got mine when they were cheap)
elsewise work with whacha got

Jim


jim
 

Ha, well, you indeed may be correct ..Since I didn't save my invoices, (maybe I didn't want to remember) and bought pieces and bits as they arrived I never kept track of exactly how much I spent ( SN 19xx ..Late-comer)

Jim


On Saturday, September 22, 2018, 4:36:38 AM PDT, Michael Maiorana <zfreak@...> wrote:


Jim,
Your post made me curious. I have an early K2 that I bought and built in 1999. Looking at my invoice, and adjusting for inflation, the K2 is about $100 cheaper today (100 2018 US dollars). 
Inflation is a thief. 
Mike M.
KU4QO

On Fri, Sep 21, 2018 at 2:55 PM, jim via Groups.Io <ab7vf@...> wrote:
I also too have the K2 manual for my unit ...Looking a "little further in front" you will notice 5 ea. band-pass filters ...Not dc to daylight transmit path ... if you want K2 performance, buy a K2 @ $800 to $1200 ...(glad I got mine when they were cheap)
elsewise work with whacha got

Jim


 

What probe are you using?

Raj

At 22/09/2018, you wrote:
This ...I measured 0.6 volts P-P on the junction or R102/R103 ...thats about 0 dbm ...Not enough for a 'level 7 ' mixer (more nearly a sine wave than a square wave or kinda a sineish square wave)

Jim

On Friday, September 21, 2018, 9:21:53 PM PDT, Jerry Gaffke via Groups.Io <jgaffke=yahoo.com@groups.io> wrote:


Howard,




The Si5351 is having a hard time dealing with the 50 ohm 6dB pads it's trying to drive,
but otherwise doing pretty good.
Jerry, KE7ER


RCBoatGuy
 

As I recall, several here have posted that the LO input to the DBM should be about 7 dBm.  That agrees with the following DBM tutorial I found online:

https://www.radio-electronics.com/info/rf-technology-design/mixers/double-balanced-mixer-tutorial.php

However, I also recall someone here mentioning that they measured the LO input as only a little over 10dB higher than the RF input during transmit.  The above tutorial recommends that the LO be about 20 dBm higher than the RF input.

What is the consensus here?  Is having the LO input to the dBM only 10 dBm higher than the RF input during transmit OK, or is the tutorial correct about needing a 20 dBM delta between the two for the best results?

73,

Carl, K0MWC