The data networking and telecom markets are driving the
bleeding edge of bit rates for high-speed digital interfaces. However,
digital-signal-processing systems in the medical imaging, wireless
infrastructure, industrial, and defense industries are experiencing an
I/O gap between the volume of data sourced by their data-acquisition
analog front ends (AFEs) and the capability of digital-signalprocessing
elements to sink the data.
In most cases, data-acquisition AFEs interface directly to FPGAs,
or indirectly to CPUs, graphical processing units, or Cell processors
through backplanes like PCI Express. According the market research
firm Databeans, the total volume of data being generated by data
converters will increase by a factor of four through 2012 (see the
figure). But, the line rates of interfaces on backplanes and FPGAs
hasn’t kept pace with Moore’s Law. Signal-compression technology
presents a solution to bridge this gap.
For signal-compression technology to address this data bandwidth
gap, it must be transparent to signal type to provide compression
results across a broad range of applications and signals.
Well-known compression technologies for video and audio, such
as MPEG2/4 or MP3, are signal-specific. And, unlike sampled data
compression, they can rely on human physiological limitations to
achieve highly lossy compression ratios.
For other industrial, scientific, and medical applications, though,
signal compression should support a lossless mode to simplify
integration into these systems without repeating any system noise
analysis. In addition, since most sampled data systems operate on
noisy signals, near lossless compression modes can provide higher
compression ratios if the noise introduced by the compression
algorithm is spectrally white and if the algorithm can control the
amount of noise introduced.
Furthermore, the compression algorithm should also offer a mode
where the compression ratio can be prescribed to simplify transmission
across fixed-bandwidth backplane and I/O interfaces. Clearly,
the value of signal-compression technology within a system is
greater when the compression is closer to the analog domain, and
when decompression is closer to the software domain, because
bandwidth increases and cost is saved throughout the hardware
signal chain. Two example applications illustrate the benefits of
signal compression in data converters.
First, in ultrasound machines, each ultrasound probe can contain
more than a hundred, or in the case of advanced 4D machines, over
a thousand transducer elements, each connected to a high-speed,
12-bit analog-to-digital converter (ADC) sampling at up to 65
Msamples/s. For a 256-element machine, this creates 200 Gbits/s of
backplane bandwidth and 512 pins between the AFE and the ultrasound
beamforming array. With 3:1 compression, the backplane
bandwidth and number of pins are both reduced by 66%, greatly
reducing the complexity and I/O power within the console.
Second, fourth-generation wireless basestations like WiMAX
and Long Term Evolution (LTE) use fiber-optic connections
between the radio electronics at the top of the tower and the
baseband processor at the bottom. As systems move from 3G to
4G, the number of antennas increases from two to four or eight
to support multiple-input multiple-output (MIMO) or smart
antenna technologies.
In addition, the channel bandwidths increase from 5 MHz for
W-CDMA to 10 or 20 MHz for LTE. This combination of factors
drives fiber-optic bandwidth requirements to the point where the
fiber-optic transceivers can cost more than the power amplifiers!
Signal compression can enable 4G basestations to reuse the same
fiber-optic infrastructure as existing 3G systems.
One such signal-compression technology is Samplify’s Prism,
which is now available in what the company says is the first compressing
ADC. The ADC, dubbed the SAM1610, provides 16 input
channels, each operating at 65 Msamples/s with 12 bits of resolution.
As a result, it can address the data bandwidth gap in ultrasound
equipment and 4G wireless infrastructure.
With decompression cores available for leading FPGA families,
as well as Intel-compatible CPUs, Cell processors, and graphic processing
units, the benefits of signal compression are now available
end to end throughout the hardware signal chain for these systems.
See associated figure