Who Wakes the Bugler

by taratuta

Category: Documents





Who Wakes the Bugler
Cart Battfes
10. Who Wakes the Bugler?
Introduction: T-Coils in Oscilloscope Vertical Systems
Few engineers realize the level of design skill and the care that is needed
to produce an oscilloscope, the tool that the industry uses and trusts. To
be really effective, the analog portion of a vertical channel of the oscilloscope should have a bandwidth greater than the bandwidth of the circuit
being probed, and the transient response should be near perfect. A vertical amplifier designer is totally engrossed in the quest for this unnatural
fast-and-perfect step-response. The question becomes, "How do 'scope
designers make vertical amplifier circuits both faster and cleaner than the
circuits being probed?" After all, the designers of both circuits basically
have the same technology available.
One of many skillful tricks has been the application of precise, special forms of the T-coil section. I'll discuss these T-coil applications in
Tektronix oscilloscopes from a personal and a historical perspective, and
also from the viewpoint of an oscilloscope vertical amplifier designer.
Two separate stand-alone pages contain "cookbook" design formulas,
response functions, and related observations.
The T-coil section is one of the most fun, amazing, demanding, capable, and versatile circuits I have encountered in 'scopes. Special forms
Figure 10-1.
The T-coil Section.
Who Wakes the Bugler?
of this basic circuit block are used with precision and finesse to do the
Peak capacitive loads
Peak amplifier interstages
Form "loop-thru" circuits
Equalize nonlinear phase
Transform capacitive terminations to resistive terminations
Form distributed deflectors in cathode ray tubes
Form artificial delay line sections
Form distributed amplifier sections
I have successfully used T-coils in all of these applications except the
last two. Recently, however, some successful designers from the *40s and
'50s shared their experiences with those two applications.
Over My Head
While on a camping trip in Oregon in 1961,1 stopped at Tektronix and received an
interview and a job offer the same day. Tektronix wanted me. They were at a stage
where they needed to exploit transistors to build fast, high-performance 'scopes. I
had designed a 300MHz transistor amplifier while working at Sylvania. In 1961,
that type of experience was a rare commodity. Actually, I had designed a wideband 300MHz IF amplifier that only achieved 200MHz. What we (Sylvania) used
was a design that my technician came up with that made 300MHz. So I arrived at
this premier oscilloscope company feeling somewhat of a fraud. I was more than
just a bit intimidated by the Tektronix reputation and the distributed amplifiers and
artificial delay lines and all that "stuff" that really worked, The voltage dynamic
range, the transient response cleanliness, and DC response requirements for a
vertical output amplifier made my low-power, 50 Ohm, 300MHz IF amplifier seem
like child's play. Naturally, I was thrown immediately into the job of designing highbandwidth oscilloscope transistor vertical-output amplifiers. I felt tike a private,
fresh out of basic training, on the front lines in a war.
The Two Principles of Inductive Peaking
The primary and most obvious use of a T-coil section is to peak the frequency response (improve the bandwidth, decrease the risetime) of a
capacitance load. Inductances, in general, accomplish this through the
action of two principles.
Principle Number One: Separate, in Time, the Charging of Capacitances
The coaxial cable depicts a limiting case of Principle Number One. A
coaxial cable driven from a matched-source impedance has a very fast
risetime. The source has finite resistance and the cable has some total
capacitance. If the cable capacitance and inductance are uniformly distrib-
Cari Battjes
^,P E A K I N G
* BJU F I C I AJ.J_0 E L A Y
{source or load)
Figure 10-2.
The Versatile T-coil.
uted and the cable is situated in the proper impedance environment, the
bandwidth is » l/2itRCcaWe and the risetime « 2.2 RCcable. The distributed inductance in the line has worked with the distributed capacitance to
spread out, in time, the charging of this capacitance. A pi-section LC
filter could also demonstrate Principle Number One, as could a distributed amplifier.
Who Wakes the Bugler?
Figure 10-3. Separate, In Time, the Charging of Capacitances,
Peaking Principle 1
C — C cable
Principle Number Two: Don't Waste Current Feeding a Resistor When a
Capacitor Needs to Be Charged In Figure 10-4 a helpful elf mans the
normally closed switch in series with the resistor. When a current step
occurs, the elf opens the switch for RC seconds, allowing the capacitor to
take the full current. After RC seconds, the capacitor has charged to a
voltage equal to IR, The elf then closes the switch, allowing the current
Figure 10-4.
Don't Waste Time Feeding a Resistor When a Capacitor Needs to be Charged.
Peaking Principle 2
Carl Battles
to feed the resistor, also producing a voltage equal to IR. No current is
wasted in the resistor while the capacitor is charging.
A current step applied to the constant-resistance bridged T-coil yields
the same capacitor voltage risetime, 0.8 RC, as the elf circuit. In both
cases, during the rise of voltage on the capacitor, the voltage waveform
on the termination resistor is negative, zero, or at least low. Without the
helpful elf, or without the T-coil, the risetime would have been 2.2 RC,
With these risetime enhancers, the risetime is lowered to 0.8 RC. This is
a risetime improvement factor of 2.75. If there are two or more capacitor
lumps, Principle Number One can combine with Principle Number Two
to. obtain even higher risetime improvement factors.
When both principles are working optimally, reflections, overshoot,
and ringing are avoided or controlled. This is a matter of control of energy flow in and out of the T-coil section reactances. A T-coil needs to be
tuned or tolerated. In the constant-resistance T-coil section, given a load
capacitance, there is only one set of values for the inductance, mutual
inductance, and bridging capacitance which will satisfy one set of specifications of the driving point resistance (may imply reflection coefficient)
and desired damping factor (relates to step response overshoot).
T-Coils Peaking Capacitance Loads
A cathode ray tube (CRT) electrostatic deflection plate pair is considered
a pure capacitance load. In the '50s and '60s, T-coils were often used in
deflection plate drive circuits. Usually a pentode-type tube was used as
the driver, rather than a transistor, because of the large voltage swing
required. The pentode output looked like a eapacitive high-impedance
source. A common technique was to employ series peaking of the driver
capacitance, cascaded with T-coiled CRT deflection plate capacitance.
The 10-MHz mtrnnix 3A6
The 3A6 vertical deflection amplifier works really hard. The 3A6 plug-in was designed to operate in the 560 series mainframes, where the plyg«s drove the
CRT deflection plates directly. The deflection sensitivity was poor {20 volts per
division) and the capacitance was high. To cover the display serein linearly and
allow sufficient overscan, the output beam power tube on each side had to traverse at least 80 volts. The T-coils on the 3A6 made the bandwidth and dynamic
range possible without burning up the large output vacuum tubes.
A RealT-Coil Response
A vertical-output deflection-amplifier designer has a unique situation—
the amplifier output is on the screen—no other monitor is needed. This
is the case with the 3A6 circuit shown here. The input test signal is clean
and fast. The frequency and step response of the entire vertical system
is dominated by the "tuning" of the T-coil L384 and its opposite-side
Who Wakes the Bugler?
Figure 10-5.
Step Response Waveforms 3A6 T-coil Peaking.
Carl Sanies
damping factor
LT = 2L + 2M
k=coupling coefficient
=— and
A/ =
and CB =
the Constant Resistance Property
2 r(l-k}R^CLs
a Quadratic (2 pole) Response at
4(1 +k)
RCs| (l~k)R2C2s2
an ALL PASS response atv3
k- 0.0 (high frequency DELAY BOOST)
V2 step response overshoot
The.nfederivefl. ^-cojlf arise from in-derived filter theory. They do not have the constant-resistance
property. The total inductance = R C. They have no bridging capacitance. They do not have a simple
quadratic (2 pole) response. The value of "m" implies a coupling coefficient k = ——
Figure 10-6.
Fact Sheet on Constant Resistance T-coils.
Who Wakes the Bugler?
counterpart. The bottom picture shows the response when the coils (L384
and its mate) were disabled. (All three terminals of each coil were
shorted together.) This reveals that, without the coils, the response looks
very much like a single-time-constant response. The middle picture illustrates the progression of tuning after the shorts are removed. The powdered iron slugs in the coil forms are adjusted to optimize the response,
The top picture shows the best response. The 10-to-90% risetime of the
beginning waveform is 75 nanoseconds, and in the final waveform it
drops to 28 nanoseconds. This is a ratio of risetimes of 2,6—near the
theoretical bandwidth improvement factor of 2.74. The final waveform
has peak-to-peak aberrations of 2%.
The total capacitance at the deflector node includes the deflection
plates, the wires to the plates, the beam power tube plate capacitance, the
wiring and coil body capacitance, the plug-in connector capacitance, the
mounting point capacitances, the chassis feedthrough capacitance, the
resistor capacitance, and possibly virtual capacitance looking back into
the tube. We can solve for the equivalent net capacitance per side by
working back from the 75nsec risetime and the 1.5k load resistance. This
yields about 23pF per side. Although each coil is one solenoidal winding,
it actually performs as two coils. The coil end connected to the tube plate
works as a series peaking coil, and the remainder as the actual T-coil.
L344, which is also a T-coil, appears upstream in the 3A6 schematic
fragment. Notice that the plate feeds the center tap of this coil. This is an
application of reciprocity (Look in your old circuit textbook!). If the
driving device output capacitance is significantly greater than the load
capacitance, it may be appropriate to use this connection.
Distributed Amplifiers in Oscilloscopes
The idea of a distributed amplifier goes back to a British "Patent Specification'' by
W.S. Percival in 1936. In August 1948, Ginzton, Hewlett, Jasberg, and Noe published a classic paper on distributed amplifiers in the "Proceedings of IRE." At about
the same time, Bill Hewlett (yes, of HP) and Logan Belleville (of Tektronix) met at
Yaws Restaurant in Portland. Bill Hewlett described the new distributed amplifier
concepts (yes, he "penciled out" the idea on a napkin!). In 1948, from August
through October, Howard Vollum and Richard Rhiger built a distributed amplifier
under a government contract. This amplifier was intended for use in a high-resolution ground radar. It had about a 6nsec risetime and a hefty output swing. In order
to measure the new amplifier's performance, Vollum and Rhiger had outboarded it
on the side of an early 511 'scope, directly feeding the deflectors.
It soon became clear that what the government and industry really needed
was a very fast oscilloscope. I am not sure of the details or sequence of events,
but Tektronix—Howard Vollum's two-year-old company—was making history.
Vollum, Belleville, and Rhiger developed the 50MHz 517 oscilloscope, an oscilloscope with a distributed amplifier in the vertical deflection path. Vollum and
Belleville had successfully refined the distributed amplifier enough to satisfy this
oscilloscope vertical amplifier application. The product was successful and order
Carl Battjes
fifjiri 10-7,
1948 Experiment—
rates exceeded Tek's ability to manufacture. Logan left Tektronix in the early '50s
and Vollum and Rhiger were left managing this new big company. John Kobbe,
Cliff Moufton, and Bill Polits, as well as other key electrical circuit designers, took
up where Vollum, Belleville, and Rhiger had left off. Other distributed amplifiers
were designed for other 'scopes during the '50s, including the 540 series at
30MHz and the 580 series at 100MHz.
Manufacturing Distributed Amplifier Oscilloscopes
The whole idea of using a distributed amplifier as an oscilloscope vertical
amplifier is rather incredible to me. Obtaining a very fast, clean step response is a hard job. When T-coils are employed* the job is even harder.
When they are employed wholesale, as in a distributed amplifier, they are
"fussy squared or tripled." The tuning of an oscilloscope distributed amplifier and/or an artificial delay line is tricky. Tuning is done in the time
domain, with clues about where and in which direction to adjust, coming
from observations of the "glitches" in the step response. If the use of a
distributed amplifier in the vertical channel of an oscilloscope was proposed in today's business climate, it would be declared "unmanufacturable." It would never see the light of day. However, the Tektronix boom
expansion in the '50s occurred largely through the development, manufacture, and sale of distributed amplifier 'scopes.
The 100MHz 580 series was the last use of distributed amplifiers in
Tektronix 'scope vertical systems. Dual triodes, low cathode connection
inductance, cross-coupled capacitance neutralization, and distributed
deflectors in the CRT helped to achieve this higher bandwidth.
istributed Amplifier Vertical Output.
Carl Battjes
Distributed Deflector for a Cathode Ray Tube
In 1961, Cliff Moulton's IGHz 519 'scope led the bandwidth race. This
instrument had no vertical amplifier. The input was connected to a
125-ohm transmission line which directly fed a single-ended distributed
deflection system. Schematics in Figures 10-8 and 10-9 show somewhat
pictorially what a distributed deflector looks like. The 519 deflector is not
shown. Within the CRT envelope was a meander line distributed deflection plate. Tuning capacitors were located at the sharp bends of the meander line. The line was first tuned as a mechanical assembly and later
incorporated into the CRT envelope.
Terminated distributed deflector structures create a resistive drivingpoint impedance in place of one lumped capacitance. They also synchronize the signal travel along the deflection plate to the velocity of the
electron beam speeding through the deflection plate length. If a distributed deflector is not used, deflection sensitivity is lost at high frequency
due to transit time. Relative sensitivity is
JL where f is frequency and fte is an inverse transit time function.
This is usually significant at 100MHz and above, and therefore distributed deflectors show up in 'scopes with bandwidths of 100MHz or
higher. Various ingenious structures have been used to implement distributed deflectors. All could be modeled as assemblies of T-coils. The effective electron beam deflection response is a function of all of the T-coil tap
voltages properly delayed and weighted.
Theoretical and Pragmatic Coil Proportions
The basis for the earliest T-coil designs was m-derived1 filter theory. The
delay lines and the distributed amplifier seemed to work best when the
coils were proportioned—as per the classic Jasberg-Hewlett paper2—at
m = 1.27 (coupling coefficient = 0.234). This corresponds to a coil length
slightly longer than the diameter. In the design phase, there was an intelligent juggling of coil proportions based on the preshoot-overshoot
behavior of the amplifier or delay line. The trial addition of bridging
capacitance invariably led to increased step response aberrations.
m-derived filters were outcomes of image-parameter filter theory of the past. The parameter "m"
determined the shape of the amplitude and phase response. "m"=1.27 approximated flat delay
response. Filters could not be exactly designed, using this theory, because the required termination was not realizable.
This classic paper described both the m-derived T-coil section and, very briefly, the constantresistance T-coil section. The use of these sections in distributed amplifiers was the main issue
and nothing was mentioned of other uses.
Who Wakes the Bugler?
In contrast with the artificial delay lines and the distributed amplifiers,
the individual peaking applications usually needed a coil with more coupling (k = 0.4 to 0.5), which was realized by a coil shorter than its diameter. When the coil value is near or below 100 nanohenries, the goal is
then to get as much coupling as possible so that the lead inductance of
the center tap connection can be overcome. Flat pancake or sandwich
coils of thin PC board material, thin films, or thick films are used to
achieve high coupling.
The Importance of Stray Capacitance in T-Coils
The stray interwinding capacitance of a T-coil can be crudely modeled by
one bridging capacitance Cbs across the whole coil. It is defined by the
coil self-resonance frequency "fres."
where Lj is the coil total inductance. If CB is the required bridging capacitance for constant-resistance proportions, then Cx=Cb-Cbs needs to be
added. This is an effective working approximation. The recent coils built
for high-frequency 50 Ohm circuits usually need additional bridging capacitance. On the other hand, the old nominally m-derived circuits never
needed any added bridging capacitance. They were high-impedance circuits with very large coils and probably had enough effective bridging
from the stray interwinding capacitance. They were probably constantresistance coils in disguise. Capacitance to ground of the coil body is always a significant factor also.
Interstage Peaking
The Tektronix L and K units of the '50s were good examples of interstage T-coil peaking. The T-coils were used to peak, not the preamp input
or the output, but in the middle of the amplifier. The interstage bandwidth
was boosted well above the
gain 2nCt«tai
gain 2nCsubtotai
The individual pre-amp bandwidths are 60MHz. This is amazing because the effective f t of the tubes was only 200MHz or so. Both inductive
peaking and f t doubling techniques were needed to "hot rod" these plugins to this bandwidth.
Carl Battjes
T-Coils in Transistor Interstages
The 150MHz 454 evolved from the 50MHz 453 oscilloscope by adding distributed
deflection plates to the cathode ray tube and, among other things, using a new
output amplifier. This amplifier employed T-coil peaking in the interstages. The Tcoil design was based on a lossless virtual capacitance, a very big approximation.
This virtual capacitance at the base was dominated by the transformation of the
emitter feedback admittance into the base. The emitter feedback cascode connection made two transistors function more like a pentode. The initial use of transistors in the early '60s showed us that, most of the time, vacuum tube techniques
didn't work with "those blasted transistors.'1 After all, vacuum tubes had a physical
capacitance that was measurable on an "off"tube; transistors had this 'Virtual
capacitance thing/! The conventional thinking in the design groups atTek in the
early and mid '60s was that inductive peaking and transistor high-fidelity pulse
amplifiers were not compatible. Despite this, the "toils and transistors did work,
the 454 worked and the 454 was a "cash cow" for Tektronix for several years.
Since then, ICs have displaced discrete transistors and the 'scope bandwidths
translated upwards, with and without T-coils. The fastest amplifiers, however, are
always produced with the aid of some T-coil configuration.
Tektronix 454
Amplifier and
Interstage T-coil,
Who Wakes the Bugler?
Phase Compensation withT-Coils
The portable 453 needed a compact delay line for the vertical system that
didn't require tuning. Kobbe had designed and developed a balancedcounterwound delay line for the 580 series of 'scopes. We made it still
smaller. This delay line worked well at 50MHz, and had reasonably low
loss at 150MHz. Unfortunately, the step response revealed a preshoot
problem. The explanation in the frequency domain is nonlinear phase
response. High-frequency delay was insufficient, and one could see it as
preshoot in the step response. Three sections of a constant-resistancebalanced T-coil structure added enough high-frequency delay to clean up
the preshoot, and even speed the risetime by moving high frequencies
into their "proper time slot." T-coil sections can provide delay boost at
high frequencies if the T-coil section is proportioned differently from that
of the peaking application. A negative value for "k" is usually appropriate and is realized by adding a separate inductor in the common leg.
Integrated Circuits
In the late '60s, when the 454A was being developed, George Wilson, head of
the new Tektronix Integrated Circuits Group at that time, wanted to promote the
design of an integrated circuit vertical amplifier. I rebuffed him, saying, "We can
never use ICs in vertical amplifiers because they have too much substrate capacitance, too much collector resistance, and too low an ft." I was correct at the time,
but dead wrong in the long run. In the 70s, Tektronix pushed 1C development in
parallel with the high-bandwidth 7000 series oscilloscopes.
Figure 10-10.
Insufficient HighFrequency Delay.
2 , 2P
3 . 3 P
3 .3
1 . 8 P
3 . 3 P
3 . 3P
Carl Battjes
1 stopped my slide into obsolescence in 1971 by doing a little downward mobility. I left the small portable oscilloscope group I headed, and joined George Wilson
in the 1C group as a designer. This foresight on my part was most uncharacteristic.
rCoiis with Integrated Circuit Vertical Amplifiers
The initial use of integrated circuits in the vertical amplifiers of Tektronix
'scopes supplied a huge bandwidth boost, but not just because of the high
ft. New processes included thin film resistors that allowed designers to
put the small value emitter feedback resistors on the chip, thus eliminating the connection inductance in the emitters of transistors. That emitter
inductance had made a brick wall limit in bandwidth for discrete transistor amplifiers. That wall was pretty steep, starting in the 150-200MHz
area. In order to have flat, non ripple, frequency response at VHP and
UHF, the separately packaged vertical amplifier stages needed to operate
in a terminated transmission line environment. T-coils were vital to
achieve this environment. Thor Hallen derived formulas for a minimum
VSWR T-coil. Packaging and bond wire layout made constant-resistance
T-coil design impossible. Hallen's T-coil incorporated and enhanced the
base connection inductance. The Tektronix 7904 achieved 500MHz
bandwidth by using all of the above, along with 3GHz transistors and
an ft-doubler amplifier circuit configuration.
In 1979, the IGHz 7104 employed many of the 7904 techniques but,
in addition, had 8GHz f, transistors, thin film conductors on substrates,
and a package design having transmission line interconnects. It also had
a much more sensitive cathode ray tube. Robert Ross had earlier developed formulas for a constant-resistance T-coil to drive a non-pure capacitor (a series capacitor-resistance combination). John Addis and Winthrop
Gross made use of the Ross type T-coils (patterned with the thin film
conductor) to successfully peak the stages and terminate the inter-chip
transmission lines.
I have lumped Thor Hallen's and Bob Ross's T-coils together in a class
I call "lossy capacitor T-coils."
Dual Channel Hybrid withT-Coils
In 1988, the digitizing IGHz Tektronix 11402 was introduced. A fast
real-time cathode ray tube deflection amplifier was no longer needed.
T-coils were employed, however, in the 11A72 dual-channel plug-in preamp hybrid (Figure 10-12), where all of the two-channel analog signal
processing took place. The T-coils peaked frequency response and minimized input reflections in the 50 Ohm input system. As in the 7904
'scope, Hallen used a design technique for the T-coils that minimized
VSWR. To realize this schematic, a T-coil was needed which had
Who Wakes the Bugler?
Two Types of Lossy Capacitor T-coils
,, v3
> RL
A>2_ _1
(3 = damping/actor of quadratic response
The Constant-Resistance property
Two Pole Response
For the Hallen and the Ross T-coils
As RB gets bigger .the input coil inductance
gets smaller.
With a finite RB, the response at RL is not
r =
Figure 10-11.
Two Types of Lossy
Capacitor T-coils.
ReCcTt(Rc+2RB) _ 2RBTt+ReRcCc
Tt+ReCc+RcCc Ll+L2
Car! Battles
enough mutual inductance to cancel the bond wire inductance that
would-be in series with its center tap. The remaining net branch inductances then had to match Hallen's values. To guide the physical layout
of this coil, I used a three-dimensional inductance calculation program.
This program was used iteratively. The two "G" patterns on the multilayer thick film hybrid are the top layer of these input T-coils. The major
dimension of these coils is 0.05 inches. In between the chips are coils
which "tune out" the collector capacitance of the transistor of each output channel. These coils are formed by multiple-layer runs and bond
wire "loopbacks."
Conspicuous by its absence is a discussion of wideband amplifier configurations and how they operate. I have referred to f,-doublers and current
doublers without explanation. I had to really restrain myself to avoid that
topic for the sake of brevity. The ultimate bandwidth limit of high-fidelity
pulse amplifiers depends on the power gain capability (expressed by an
fMAx» f°r example) of the devices, and the power gain requirements of the
amplifier. To approach this ultimate goal requires the sophisticated use of
inductors to shape the response. For bipolar transistors, the ft-doubler
configurations and single-stage feedback amplifiers, combined with inductive peaking, do a very good job.
I hope this chapter has raised your curiosity about the circuit applications of the T-coil section. I have not written this chapter like a textbook
Figure 10-12.
witi Thick Film
Who Wakes the Bugler?
and I am hoping that my assertions and derivation results are challenged
by the reader. To get really radical, breadboard a real circuit! A less fun
but easier way to verify circuit behavior is via SPICE or a similar simulator program. Keep in mind, while you are doing this, that most of the
very early design took place without digital computer simulators.
Frequency- and impedance-scaled simulations took place though, with
physical analog models.
I'm grateful to the many knowledgeable folks who talked with me
recently and added considerable information, both technical and historical. These included Gene Andrews, Phil Crosby, Logan Belleville, Dean
Kidd, John Kobbe, Jim Lamb, Cliff Moulton, Oscar Olson, Ron Olson,
and Richard Rhiger. If this chapter has errors, however, don't blame these
guys; any mistakes are my own.
Bob Ross and Thor Hallen have been sources of insight on these topics over many years and have been ruthless in their rigorous analyses,
helping me in my work immensely.
Finally, I leave you with my mother's and Socrates' advice,
"Moderation in all things." Might I add, "Just do it!" If these Tek guys
had waited for proper models of all known effects and proper theory
before doing something, we would still be waiting. Everything can be
tidied up in hindsight but, in fact, the real circuits in the real products are
often more complicated than our simple schematics and were realized by
a lot of theory, intuition, and especially smart, hard, and sometimes long
work. I am proud of all of this heritage and the small part I played in it.
Fly UP