[evla-sw-discuss] Re: Guidelines for EVLA MIB and digital electronics

[Nick Peereboom]

Jim,
I agree with addressing self-interference early in design. It is relatively
inexpensive to design-in basic noise reduction techniques at the beginning
of design, leaving room for noise-reduction add-ons.  Noise reduction
becomes far more expensive once boards have been designed, laid out, and
integrated into a system.

The common textbook solution of "capacitive decoupling" is inadequate for
many applications. I suggest we use these inter-related noise reduction
techniques:

- use lots of cheap surface mount three-terminal ferrites (Murata, 0805
outlines)
- use picket-fence plated-through vias (suggest 22.5 mil diameter) to
isolate independent circuits in rectangular regions.
- when necessary add a metal shell (sheet brass is best) over picket-fence
vias. Solder in-place to make a small local shielded volume
- use dedicated local ultra-low noise voltage regulators in tiny SOT-23-5
packages (LinearTechnology, example: model LT1761)
- select proper capacitor dielectric types and values especially around
voltage regulators (see Linear Technology voltage regulator data sheets)
and as DC blocks (see Agilent "AppCAD" software, active devices section)
- minimize distance between dedicated voltage regulators and the device
being powered
- use surface mount low-pass and band-pass filters (Synergy) around active
analog devices
- use low noise surface mount op-amps (Analog Devices) instead of bipolar
amplifiers for gain at frequencies less than 100 MHz. Low noise low cost
op-amps are available with far better intermodulation performance than
bipolar amplifiers. Depending on application, the op-amp gain can be
changed in real-time during prototyping. The designer can tailor the
bandwidth of the op-amp circuit to be what he or she wants to exclude
undesireable higher frequency signals like harmonics. The designer is free
to optimize op-amp interstage impedance matching and minimize return loss.
- use Spectrum Control #4-40 thread bulkhead feedthroughs for power, signal
and ground lines in Tee-filtered configuration
- use Spectrum Control DB9, 15, 25, 37, and 50-pin individually
Tee-filtered lines in D-shell connectors
- design schematics and generate layouts for isolation of independent
circuits. Review schematic and layout with colleagues before fabrication.
- use a low dielectric constant PCB prepregs like WL Gore Speedboard.
"SpeedBoard" is a material designed to replace and improve upon old "FR4".
The low dielectric constant in SpeedBoard greatly reduces cross-talk and
capacitive loading for high speed digital boards at datarates DC through 10
GHz. Speedboard material cost is marginally more than for FR4, but it is
easier to process so the total fabrication cost is about the same.
- physically separate digital circuits, serial data and serial control
lines and digital buses from microwave, HF, VHF and low frequency analog
circuits, especially in plug-in assignments and design
- control microwave devices with true "TTL" levels. While many microwave
devices are advertised as "CMOS compatible", a true TTL level with the
relatively high current sourcing capability of TTL is needed to reliably
drive microwave devices.
- for every digital TTL line, layout a parallel "return line" with a direct
path to the ground on the driver chip. Use SMT ferrites in each return line.
- implement digital control of microwave and low frequency circuits in a
"hands off" manner. By "hands off " I mean connect to microwave plug-ins
with digital TTL parallel lines (signal/return wire pairs) through three
terminal SMT ferrite filters, through filtered backshells, use a TTL
receiver chip heavily filtered with SMT 0805 footprint ferrites, and
finally drive the microwave device thru a SMT ferrite filter. This sequence
permits millisecond-by-millisecond control of the digital aspect of the
microwave device while avoiding the introduction of digital spurious noise
in 0.5 to 1000 MHz onto analog lines.
- use 0.010" thick sheets of microwave absorber when necessary

While there are many useful noise reduction techniques, this list is
current. Some of the techniques actually convert microwave and lower
frequency noise power into harmless thermal  energy. 

~Nick

At 09:05 AM 1/15/2002 Tuesday , Jim Jackson wrote:
>As the hardware systems engineer, I feel it is time to issue some
>guidelines relating to the design of the EVLA Monitor Control Interface and
>other digital electronics that will be used in the antenna and central LO
>system modules.  Based on the emails I'm receiving and comments I'm hearing
>in meetings, I becomming increasingly concerned that the most important
>technical issue with these modules is not being sufficiently considered.
>That issue is RFI.  Here are some guidelines for the design of the MIB
>intended to minimize thge RFI risk in the system.
>
>1.  In normal operation, all code to be executed should reside in on-chip
>memory on the microprocessor / microcontroller IC.  High speed external
>buses (especially for program execution) can be a major source of noise
>emission and should be avoided.  Loading the firmware into on-chip SRAM
>from external non-volatile memory at startup is acceptable.
>
>2. Only the minimal required amount of processing should be being performed
>inside these modules or even in the vertex room of the antennas. The more
>processing that can be moved to the pedestal room or CEB (preferred) and
>away from the front end and LO/IF systems, the better.  
>
>3. M&C communications with each module or subsystems must be via fiber
>optic cable.  The fiber optic connectors on the modules must be adequately
>shielded. The Diamond E-2000 series blindmate backplane connector is being
>considered though we will need to find a way to shield them 
>
>4.  Many new peripheral devices such as A/D and D/A converters use a 2 or 3
>wire serial interface (I2C or SPI respectively) for communication.   These
>interfaces may be used (in fact, for maximum design flexibility, I would
>encourage it), however, the interfaces should be operated at as low a clock
>rate as possible to perform the required functions. They also must be
>properly terminated. It may also be possible to control the rise times of
>these signals to reduce emission. 
>
>5.  Digital electronics in the MIB (and anywhere else in the modules)
>should be run at the lowest clock rates possible to still perform the
>requred task.  Wherever possible, digital electronics should be clocked by
>indepenent free-running clock sources that are not sychnronized to the
>Master LO/Reference system. 
>
>6.  Digital electronics should be operated at the lowest possible voltages
>(I would prefer that 3.3VDC logic be used instead of 5VDC logic wherever
>possible).  Digital and analog electronics should be operated from separate
>voltage regulators (though these regulators will almost certainly be
>connected to a common source). Devices which use even lower core voltages
>(like 2.5, 1.8, or 1.5) are even better. 
>
>7.  Circuit boards should be designed using low noise design techniques.
>Also, where possible look at controlled rise time techniques on digital
>signals.  Fast rise time digital signals distributed between modules and
>systems should always be implemented using a low impeadance differential
>logic like RS422 or LVDS.  Low rate digital signals can be filtered to
>control risetimes - they can be single ended or differential.
>
>Based on these guidelines, I do not think devices like the Compulab 586
>board being proposed for the correlator are suitable solutions for the MIB.
>They use external program memory, operate at far higher clock rates (and
>provide far more processing power) than should be required to perform the
>tasks required.  The Compulab board also does not appear to contain an SPI
>interface that will be required to communicate with many modern A/D and D/A
>converters - though other boards of this type may include this feature.
>These devices do look like an excellent match for the correlator and
>possibly a few other select applications where more computing horsepower is
>required.
>
>The EC-1 device being proposed by some looks like a reasonable solution.
>My only reservation with that device is that it is from a relatively small
>relatively unknown company and may have an uncertain future so let's make
>sure we exhaust other possibilities before we use it. We do need to try to
>pick a device that will remain available, at least through the design and
>development phase of the project (or up to the point where we have the
>money to buy a lifetime supply of them). 
>
>Jim Jackson
Nick Peereboom
NRAO - ALMA Program, AOC Building, PO Box "O", 1003 Lopezville Road,
Socorro New Mexico USA 87801-0387
Telephone: 505-835-7216 direct, 505-835-7000 front desk Fax: 505-835-7027 &
505-835-7024