SV: Re: SV: Re: [sdiy] BBD clock circuit

[karl dalen]

Swoshless flanging?

KD


--- jhaible at debitel.net skrev:

> > JH, how does the 1/X function work?
> > Is it doing log arithmetic?
> 
> Yes. Starting from the back end, Q1A performs an expo function.
> (collector current vs. BE voltage). It's important that you need
> a _negative_ BE voltage here. So in all the log arithmetic below, negative 
> voltages are counted as "+", and positive voltages are counted as "-".
> 
> Now starting at the front end. Consider pin5 of U1B at a fixed voltage.
> Let's asume 0V for now.
> 
> Then R3 sinks a current i3 to a virtual GND node (pin6), and the "diode"
> Q2A performs a log function. (Don't bother about constants here - they
> will all cancel in the end.) Important: As R3 is sinking, the log'ed
> voltage goes up (opamp output), and we count this as "-".
> 
> The voltage from the opamp output is now downshifted by two diode voltage
> drops (Q2B and Q1B). We count this as a "+" contribution. Important: the
> current thru these diodes is different: It's the current i1 set by R1.
> 
> Now the expo transistor sees a voltage at its base which is basically
> (without constants) -log(i3) + 2*log(i1). 
> 
> After performing the expo function, the log's dissapear, subtraction becomes
> a division, and a multiplication with 2 becomes squaring.
> So the collector current of Q1A is 
> 
> i_out = i1**2 / i3.
> 
> i3 was out input current, amd it's magically found in the denominator.
> 
> But the _real_ magic is that the same circuit performs an expo function
> from the other input of the opamp (pin5) at the same time. (Which is
> easy to show.)
> 
> I hope this was of some help, even though omitting all the constants
> may look criminal. They really all just cancel in the overall function.
> 
> I think whoever invented this circuit is a true genius. Dynacord,
> for comparison, made the 1/x function in the TAM-19 with a 3080 in
> an opamp feedback path to divide, and a thousand of trimpots.
> 
> JH.
> 
> -------------------------------------------------
> debitel.net Webmail
>