|
Synthetic Instruments and LXI
by David Poole, Aeroflex, and Bob Rennard, Agilent Technologies
Synthetic instruments may have found the common interface they require in LXI.
Synthetic instruments (SIs), built from modular components and enabled by high-speed processors and modern bus technologies, promise test users
increased functionality and flexibility, lower total cost of ownership, higher
speed operation, smaller physical footprint, and longer supportable life. One
problem that both manufacturers and customers face is the lack of a common
design standard that meets their architectural and commercial needs.
PXI and VXI modular instrumentation are candidates for this application. But
they force many constraints on the designers of SI, resulting in hybrid systems
with proprietary interfaces and custom module packaging.
SI Top-Level Requirements and Architecture
At first glance, the general requirements for SI are similar to those of
conventional rack-and-stack instruments. On further investigation, we discover
hidden requirements unique to synthetic architectures, so it is worth starting
with a quick look at general SI systems.
There have been many definitions of SIs over the years, leading to some
confusion and misunderstanding. At the most fundamental level, SIs simply are
collections of generic hardware and software modules that may be combined to
perform a variety of traditional test functions.
Recently, the SI Working Group, sponsored by the Department of Defense, provided
stability to the discussion by publishing the architectural description shown in
Figure 1. The SI Working Group was chartered to define SI component/module
specifications for military applications and promises to have a significant
effect on the course of SI development in the coming years.
|
Figure 1. General SI Architecture |
SI architectures provide the capability to assemble instruments from standard
modules individually selected to supply the required test capabilities and
specifications. Figure 1 shows the main components of a typical SI
stimulus/response system but fails to help us understand the actual
characteristics of the modules themselves.
System integrators using SI architectures must select modules capable of
supporting the desired system specification and assemble them into an
architecture with adequate data and trigger-bus performance. This modular
architecture provides many benefits in terms of cost/performance trade-offs. For
example, you would not select an expensive, extremely fast-switching local
oscillator if the final system only makes single frequency measurements.
With SI, in an ideal world, systems integrators could select from a range of
components with common interfaces but feature different capabilities and price
points when building test systems. The challenge that the SI manufacturers now
face is the definition and acceptance of a common standard that would allow this
module assembly to actually happen without the use of a hacksaw.
SI Modules and Components
So can we define general module characteristics and derive interoperability requirements? It probably is better to start with the physical
requirements since these define how modules fit together in a typical system
rack.
Some attributes are fairly clear: We would like a package that provides good RF
isolation, size flexibility, adequate cooling, and easy IO access. We would like
the overall size to be flexible to accommodate large, high-power microwave
components, and we would like power supplies to scale with module needs, such as
baseband switch modules that consume far less power than RF.
We would like to build from the lessons card-cage modular formats taught us
about the limits of tightly constrained physical enclosures that do not
accommodate large RF modules and avoid isolation problems between computer bus
standards and sensitive RF signals. An additional concern is that the
over-constrained physical requirements associated with card-cage formats
conspire to limit module availability and can drive up prices by preventing
manufacturers from leveraging investments in rack instruments into SI modules.
Electrical requirements, particularly data and control bus bandwidth and
latencies, often drive module design decisions. Baseband modules can drive
significant data between modules, particularly with systems that generate and
acquire complex modulation patterns where data rates of 20 to 50 MB/s are not
uncommon for modern military digital communications protocols.
The data and control requirements between up and down converters, local
oscillators, signal conditioning, and DUT interfaces are modest. Switch
synchronization requirements vary between the microsecond level for pin diodes
out to tens of milliseconds for mechanical switches and attenuators.
Synchronization between stimulus and response and baseband modules, historically
in the tens of nanoseconds for demanding applications, is more challenging.
This gives us a picture of a composite system where the baseband modules are
connected by a high-speed data bus with bandwidths exceeding 50 MB/s within a
larger system with data bus bandwidths in the order of 10 MB/s. Although these
requirements can vary for specific applications, we believe they are fairly
typical for the majority of SI applications and give us a starting point for the
discussion about the applicability of the LXI standard.
LXI Overview and History
In 1972, the Hewlett-Packard Interface Bus (HP-IB) was introduced as an open
standard communications bus (IEEE 488) for instrument-to-computer
communications. Later renamed general-purpose interface bus (GPIB), it was found
on nearly every instrument over the past 30 years and defined test-system
architectures. The popularity of GPIB was driven by longevity and stability; it
was a single interface standard test-system designers could count on.
In 1985, Hewlett-Packard (now Agilent Technologies), Tektronix, Wavetek,
Racal-Dana, and Colorado Data Systems introduced VME eXtensions for Instruments
(VXI), a modular instrument standard for the U.S. military. These modular
instruments became very popular in the aerospace/defense industry and
manufacturing test applications where size and throughput were important.
But time moves on, and new technologies become available. Unlike the early days
of GPIB when simple computer-to-instrument communications were the goal and few
alternatives existed, low-cost serial communications buses have emerged with
extremely high performance. These modern switched-fabric technologies such as
Infiniband, Rocket I/O, PCIExpress, and Gigabit Ethernet changed the face of
computer communications architectures.
Conventional parallel buses became expensive bottlenecks for communications
between nodes on a network. In this changing environment, test and measurement
manufacturers saw the advantages of using low-cost, high-speed serial buses for
instrument communications, releasing them from parallel-bus architectural
constraints.
Ethernet is an attractive solution. It is a stable, single ubiquitous high-speed
connectivity standard with extensive network management capabilities and media
independence.
Over the past two decades, Ethernet evolved far faster than test and
measurement-derived interfaces, yet it preserves excellent backward
compatibility and stability. In the latest Gigabit incarnation, Ethernet finally
has achieved the performance required to become the fabric of the interface
architecture required by the SI community.
In 2004, Agilent Technologies and VXI Technology founded the LAN-based
eXtensions for Instrumentation (LXI) Consortium, a not-for-profit corporation to
develop and promote an open industry standard that ensures interoperability
between local area network (LAN)-based instruments.1 With many companies
independently considering Ethernet-based instruments and SI modules for the
functional test, measurement, and data acquisition industries, defining
consistent implementation practices made sense. The founders reasoned that
independent implementation would result in poor interoperability and doom
Ethernet as a viable successor to GPIB.
LXI Functionality and Performance
LXI primarily is a functional interface specification that defines
implementation practices for Ethernet 802.3-based interface technology to ensure
interoperability between instruments with an embedded IEEE 1588 protocol to
provide the required synchronization capabilities.2 In addition, it specifies an
associated wired trigger bus to provide enhanced synchronization capabilities
for critical applications.
There are three LXI conformance classes, differing primarily in mechanical
dimensions and triggering/synchronization requirements:
• For SIs, LXI Class A is the most interesting, with deterministic timing
provided by the IEEE 1588 synchronization protocol and a high-speed wired
low-voltage differential signalling trigger bus. Since most SI modules have no
front panel or display, the user interface is defined through a web interface,
and an interchangeable virtual instruments application programming interface
(IVI API) provides communications among computers and modules. LXI mechanical
specifications ensure modules behave as good system neighbors.
• The trigger bus operates at 100 MHz, providing 1-ns accuracy with installed
achievable performance of approximately 2 ns, and can be configured in star,
daisy-chain, or hybrid architectures. The IEEE 1588 protocol uses real-time
clocks to provide deterministic network timing independent of underlying LAN
speed.
• The IEEE 1588 protocol provides timing synchronization to around 50 ns at a 2-sigma distribution over the LXI Ethernet bus. The data throughput
and latency of the LXI bus obviously depend to a large degree on the processors
and interface hardware. As for Gigabit Ethernet performance, it is worth quoting
from one of the many sources on this topic:
“Gigabit Ethernet has now become commonplace in servers and workstations. While
its signaling rate of 1 Gb/s translates to a peak bandwidth of about 120 MB/s, this performance level
is rarely achieved in practice. The standard connection-oriented protocol
carried on Ethernet is TCP/IP. Because the host CPU is generally responsible for
implementing the computer intensive TCP/IP software stack, one typically
observes a relatively large latency, in the range of 50 to 120 µs, for zero-byte
messages.”3
The magnitude of these latencies should be kept in the context of typical SI
design requirements. Most advanced SI architectures have moved to the use of
common test schedules or state machine schedules distributed to the various
active modules in the system, with the schedule synchronization being
accomplished in the 1588 protocol. This allows widely distributed components to
synchronize events in approximately 100 ns. More accurate synchronization of the
schedule events can be accomplished by the wired trigger bus.
The only time that the LXI bus latencies become significant is in the execution
of procedural events in a script-like environment. In this case, typical
programming techniques will require the host CPU to switch context to execute
the command, and there also may be other latencies involved in script
interpretation.
The context switching time for both Microsoft XP and Linux was investigated by
Dr. Bradford in Reference 4 and reported as commonly exceeding 30 µs. With any
additional interpreter latency, this and not the Ethernet bus latency is the
dominant factor in end-to-end data and event transport between modules.
SI Designs Using the LXI Standard
SI systems can easily be built with parts of the baseband architecture embedded
in a VXI or PXI cage. However, it is difficult to fully contain the RF
architecture within this same mechanical and electrical environment, especially
for high-performance microwave systems. This has led many manufacturers to
design hybrid systems using a combination of card-cage-based baseband connected
to larger RF components.
With these solutions available, why does LXI provide an attractive new standard
for these applications? The answer has to do with the trade-offs and constraints
imposed by card-cage solutions and the resultant hybrid architectures used to
avoid these limitations.
Baseband and low-frequency modules lend themselves to VXI and PXI, but it has
proven very difficult to build RF and microwave systems within the mechanical
and electrical environment presented by these standards. RF and microwave
components typically do not fit within card constraints, shielding is very
challenging, and power-hungry RF devices often tax power supplies.
The result is a lack of high-performance RF and microwave products in these
formats, forcing system integrators to develop hybrid systems with baseband
modules in a commercial cage combined with custom RF modules controlled by
proprietary data and control buses. It is not in the interest of either the
manufacturer or consumer to deal with independent implementations since they
will negate many of the module interchange and system configurability benefits that otherwise would
be inherent in these SI designs.
A standards-based system using Ethernet as the main system bus offers many
attractive advantages. Advanced network features such as peer-to-peer
communications are not available in traditional test and measurement interfaces,
and Ethernet cables and cards are standard, widely available, and essentially
free.
Power-over-Ethernet (PoE) and the IEEE 1588 protocol simplify distributed
measurement applications where large numbers of sensors are distributed around a
large DUT or distributed geographically, and IEEE 1588 time stamping simplifies
data and channel management and post-acquisition analysis. A possible
architecture for a typical stimulus/response SI test system is shown in Figure
2.
The architecture has separate stimulus and response baseband modules connected
to a server via the LXI bus and an LXI trigger bus between baseband modules. In
this design, the frequency converters provide signal conditioning.
But in practice, the task could be separated as shown in Figure 1. The DUT
interface provides interface switching and loop-back calibration functions but
also could include signal conditioning, high-power stimulus amplification, and
low noise response amplification.
This example, built around the LXI standard, easily meets the needs and exceeds
the performance of most SI systems built today. Bus bandwidths are 50 MB/s,
baseband trigger accuracies are about 2 ns over the LXI trigger bus, and up/down
converter and DUT trigger accuracies over IEEE 1588 are well below 100 ns.
Further, IEEE 1588 allows time stamping for post-acquisition analysis that
previously was not available with event-driven triggering.
LXI Packing Density, Cooling, and Power
Since LXI is primarily an interface standard, it presents few mechanical
constraints on test instruments. For SIs and applications where size is a
priority, the LXI mechanical specification defines a 1U half-rack width module
that can be tightly integrated in a stack of similar modules with cooling and
cabling interfaces designed for this type of application.
The rear panel has the LXI and trigger-bus connectors, and either AC or DC power
is allowed. The front panel contains the status lights and reset buttons, and
the units can be mounted in a 19-in. rack through the use of the intermodule and
rack-mounting ears.
Although manufacturers will want to qualify their own units for thermal
stability, initial studies by the consortium have shown that each 1U unit can
reasonably handle around 100 W without any particular design concerns, and 2U
units are good for around 400 W on a similar basis. Figure 3 illustrates this
configuration for high-density applications, with a combination of 1U and 2U
modules in a high-density stack suitable for a 19-in. rack-mount enclosure.
|
Figure 3. LXI Module Stacks for Rack-Mount Applications |
Conclusions and the Road Ahead
In SI system scenarios, LXI is an attractive alternative built upon open
Ethernet standards and free from the size constraints imposed by existing
bus-specific standards. It offers a component-based architecture capable of
meeting both the low-cost and the high-performance needs of test-system
integrators with a digital bus interface tied to modern telecommunications
hardware development.
By leveraging Ethernet, LXI allows access to state-of-the-art high-performance
processors and servers. But most of all, the emergence of a true physical and
functional component-based architecture, capable of meeting both the low-cost
and the high-performance ends of the SI spectrum, can provide benefits in the
establishment of a truly interchangeable commercial SI standard.
References
1. LXI Consortium, http://www.lxistandard.org/home
2. IEEE 1588 Standard, http://ieee1588.nist.gov/
3. Jonsson, Dr. L.E., and Magro, Dr. W.R., “Comparative Performance of
InfiniBand Architecture and Gigabit Ethernet Interconnects on Intel® Itanium® 2
Microarchitecture-Based Clusters,” Intel Americas.
4. Bradford, Dr. E.G., “High-Performance Programming Techniques on Linux and
Windows,” IBM, July 2002.
About the Authors
David Poole works for the System Division of Aeroflex as a technical fellow and
is the chairman of the LXI Physical Specifications Working Group. He has degrees
in math and physics, is a graduate of the USN Test Pilot School, and has a
background in math models, digital control, instrumentation, and test systems.
Aeroflex Systems Division, 4201 Northview Dr., Bowie, MD 21706, 410-693-2561,
e-mail: david.poole@aeroflex.com
Bob Rennard is a program manager at Agilent Technologies and president of the
LXI Consortium. He received degrees in engineering and an M.B.A. from
Northwestern University and has held a number of positions in the test and
measurement business, including marketing manager for spectrum analyzers and
signal generators. Agilent Technologies, 1400 Fountaingrove Parkway, Santa Rosa,
CA 95403, 707-577-3140, e-mail: bob_rennard@ agilent.com |
