T
Tim Williams
Guest
<upsidedown@downunder.com> wrote in message
news:3535jf1d6gcioqq6s4ombgffdm8p37prl6@4ax.com...
Usually the voltages are modest (~100V), so that adequate current can be
drawn. Low voltage, you need too much cathode area and capacitance, and
your figure of merit (Gm/(Cin+Cout), which has units of frequency) stinks.
And obviously, the FoM doesn\'t change with scale (length/area of cathode).
Though other fixed parameters (like wiring strays) do affect the FoM
in-system, so you don\'t want to use too small of a tube.
They can, but not baseband to VHF. Only a few late era planar types could
do that. (Which at the time, were more valuable doing the same ~100MHz
bandwidth around a center frequency of some GHz -- pretty good bandwidth
even today, to be fair!)
That\'s the trick. Example:
I happen to have a carton of subminiature (wire lead) 5702s.
These are your basic general purpose / RF, sharp cutoff pentode. 1W max,
5mS at 100V, 10mA.
In an accidentally-grounded-grid arrangement, these like to oscillate around
400MHz; it\'s the interelectrode ESL+Cp resonant mode I think. (This might
happen accidentally, like in a follower type Hartley oscillator, which
places its tuning capacitor between grid and ground.) They\'re perfectly
capable of operating at high frequencies; the transit time is a nanosecond
or two. (This can be measured with a very low plate load and a fast pulse
generator into the grid. The gain in this configuration is pitiful,
obviously, but you can indeed see the current changing in that time!)
The capacitances are about 4pF plate and 5pF grid *WHEN COLD*. However the
grid capacitance increases when hot -- because the electron cloud has mass,
cool huh? -- to more like 9pF. So the total in cascade is 4+9 = 13pF. Plus
circuit strays.
That gives a FoM = 5mS / (2*pi*13pF) = 61MHz, which is the no-frills (no
peaking) ballpark bandwidth, at unity gain (i.e., a 1/5mS = 200ohm plate
load).
In a practical computer, you\'d need 1/2 to 1/3 of this (even for a tiny
fanout of, well, about as much), and even with peaking (to about double it,
give or take how that works in combination with nonlinear loads like diode
gates), you\'re looking at a maximum clock rate for typical gates of maybe
20MHz.
Which, is still pretty promising; that\'s not much slower than 74HC logic,
and faster than CD4000, scarily enough. It does take about a dozen watts to
compute even fairly simple logic operations though... ECL\'s got nothing on
this!
Heh, given that those have been obsolete for a while now too -- not hard to
find though.
I wonder how much data those store. Let\'s see, 64us, at a MHz or two
bandwidth -- they had to store chroma, modulated I believe, so that\'s offset
on a center frequency? -- suggests 2Mbit/s * 64us = 128 bits. Although
that\'s not actually bits, but baud, and more bits could be encoded given
some manner of ADC/DAC system, with adequate SNR and ISI (with whatever ISI
compensation can be afforded).
Not bad for, like, a register file, or perhaps more practically, a cache
line..?
And since it\'s modulated, it needs some encoding. PAM perhaps, which would
be an easy enough way to transmit 2 bits/symbol (BPASK I guess you might
call it?). Probably it would be easy enough making a 4x4 square
constellation (4 bits/symbol) too.
Which puts it up around half a kiB, which isn\'t too shabby. And it would
probably match up well enough with a 5702-based (or other tubes of
comparable performance) machine.
Downside, now you have to \"spin the drum\" or whatever; efficient programming
gets a lot harder...
(I wonder if anyone\'s ever written a compiler/optimizer around such a
constraint? Certainly very little reason to do so today, but as an academic
exercise, maybe interesting.)
Well, I don\'t know about that. They certainly worked..when they did. How
many of them are even around anymore? And of them, how many are basically
in parts, or in need of extreme-teardown levels of maintenance? :^)
Definitely mechanical wonders; I\'ve quite enjoyed CuriousMarc\'s restoration
series:
https://www.youtube.com/watch?v=_NuvwndwYSY
And absolutely, if you have one in working order -- perfectly compatible
with the same old serial frames we\'ve always used, if at rather low baud
rates. Just insert RS-232 (or TTL, or..) to 20mA converter, and don\'t touch
the high voltages.
Tim
--
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Design
Website: https://www.seventransistorlabs.com/
news:3535jf1d6gcioqq6s4ombgffdm8p37prl6@4ax.com...
Usually the voltages are modest (~100V), so that adequate current can be
drawn. Low voltage, you need too much cathode area and capacitance, and
your figure of merit (Gm/(Cin+Cout), which has units of frequency) stinks.
And obviously, the FoM doesn\'t change with scale (length/area of cathode).
Though other fixed parameters (like wiring strays) do affect the FoM
in-system, so you don\'t want to use too small of a tube.
They can, but not baseband to VHF. Only a few late era planar types could
do that. (Which at the time, were more valuable doing the same ~100MHz
bandwidth around a center frequency of some GHz -- pretty good bandwidth
even today, to be fair!)
That\'s the trick. Example:
I happen to have a carton of subminiature (wire lead) 5702s.
These are your basic general purpose / RF, sharp cutoff pentode. 1W max,
5mS at 100V, 10mA.
In an accidentally-grounded-grid arrangement, these like to oscillate around
400MHz; it\'s the interelectrode ESL+Cp resonant mode I think. (This might
happen accidentally, like in a follower type Hartley oscillator, which
places its tuning capacitor between grid and ground.) They\'re perfectly
capable of operating at high frequencies; the transit time is a nanosecond
or two. (This can be measured with a very low plate load and a fast pulse
generator into the grid. The gain in this configuration is pitiful,
obviously, but you can indeed see the current changing in that time!)
The capacitances are about 4pF plate and 5pF grid *WHEN COLD*. However the
grid capacitance increases when hot -- because the electron cloud has mass,
cool huh? -- to more like 9pF. So the total in cascade is 4+9 = 13pF. Plus
circuit strays.
That gives a FoM = 5mS / (2*pi*13pF) = 61MHz, which is the no-frills (no
peaking) ballpark bandwidth, at unity gain (i.e., a 1/5mS = 200ohm plate
load).
In a practical computer, you\'d need 1/2 to 1/3 of this (even for a tiny
fanout of, well, about as much), and even with peaking (to about double it,
give or take how that works in combination with nonlinear loads like diode
gates), you\'re looking at a maximum clock rate for typical gates of maybe
20MHz.
Which, is still pretty promising; that\'s not much slower than 74HC logic,
and faster than CD4000, scarily enough. It does take about a dozen watts to
compute even fairly simple logic operations though... ECL\'s got nothing on
this!
Heh, given that those have been obsolete for a while now too -- not hard to
find though.
I wonder how much data those store. Let\'s see, 64us, at a MHz or two
bandwidth -- they had to store chroma, modulated I believe, so that\'s offset
on a center frequency? -- suggests 2Mbit/s * 64us = 128 bits. Although
that\'s not actually bits, but baud, and more bits could be encoded given
some manner of ADC/DAC system, with adequate SNR and ISI (with whatever ISI
compensation can be afforded).
Not bad for, like, a register file, or perhaps more practically, a cache
line..?
And since it\'s modulated, it needs some encoding. PAM perhaps, which would
be an easy enough way to transmit 2 bits/symbol (BPASK I guess you might
call it?). Probably it would be easy enough making a 4x4 square
constellation (4 bits/symbol) too.
Which puts it up around half a kiB, which isn\'t too shabby. And it would
probably match up well enough with a 5702-based (or other tubes of
comparable performance) machine.
Downside, now you have to \"spin the drum\" or whatever; efficient programming
gets a lot harder...
(I wonder if anyone\'s ever written a compiler/optimizer around such a
constraint? Certainly very little reason to do so today, but as an academic
exercise, maybe interesting.)
Well, I don\'t know about that. They certainly worked..when they did. How
many of them are even around anymore? And of them, how many are basically
in parts, or in need of extreme-teardown levels of maintenance? :^)
Definitely mechanical wonders; I\'ve quite enjoyed CuriousMarc\'s restoration
series:
https://www.youtube.com/watch?v=_NuvwndwYSY
And absolutely, if you have one in working order -- perfectly compatible
with the same old serial frames we\'ve always used, if at rather low baud
rates. Just insert RS-232 (or TTL, or..) to 20mA converter, and don\'t touch
the high voltages.
Tim
--
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Design
Website: https://www.seventransistorlabs.com/