From: Winfield Hill
Subject: Re: Prof Searle's 100MHz 30dB 1968-era 3-trannie amplifier
Date: 18 Jan 2003 19:03:32 -0800
Organization: Rowland Institute
References: <3E2358BE.firstname.lastname@example.org> <3E2A0C89.7FAB@sneakemail.com>
X-Newsreader: Direct Read News 4.20
>Winfield Hill wrote:
>> OK, I'll start by posting a simple back-of-the-envelope design. The
>> drawing below is taken from our book, figure 13.13, where we describe
>> the operating principle of the amplifier. It has two local emitter-
>> degeneration-feedback gain stages (R2/R3 and R4/R5), plus two more
>> nested two-transistor feedback loops (R6/R1 and R7/R3). Clever, huh?
>>. shunt-series pair |
>>. Art of Electronics R4
>>. page 872 V+ |
>>. | +----+---- out
>>. R2 Q2 | |
>>. | C |
>>. +------- B |
>>. | E |
>>. | | |
>>. ,--- | --- R6 --+ |
>>. | | | |
>>. | C R5 |
>>. -||-- R1 -+- B Q1 | |
>>. E gnd |
>>. | |
>>. +---||--- R7 ---'
> Golly, that sure looks familiar. I can't say for sure as it was such
> a long time ago, but let's go with it.
> As I recall, the input was driven from the 50 ohm source impedance of
> a network analyzer. I'm not sure, but I think the output drove 50 ohms
> directly. That might account for the third transistor used as an
> emitter follower, but I rather see what else you had in mind.
Yes a good possibility, although no voltage gain results.
> The compensation cap went across R7.
Hmm. Normally there isn't enough loop gain in the shunt-series pair
beasts I've designed to need such a trick.
> Should there be a bypass cap from the emitter of Q2 to ground? R5 is
> going to be quite large due to the stacked voltages, which would make
> R4 really large to obtain much gain. Bandwidth would suffer, and the
> feedback resistor R7 would start to dominate.
Yes, exactly right. Ideally one would share the gain task between the
two transistors, but I wasn't able to get the 2nd stage gain up to snuff
and a bypass (with a series resistor?) could well solve that. Ah, with
no degenerating series resistor for the bypass cap, perhaps a cap across
R7 might be needed after all, to add a fortuitous zero in the loop.