Many different processors are available to provide digital signal processing. These processors
requirement in selecting a processor. Other factors including dynamic range, arithmetic
precision, cost, and size are also important considerations when choosing a processor.
A common method used to increase total processing speed beyond that of a single processor is to
employ multiple processors operating in parallel. Assuming a given processor, by putting more
processing speeds can be achieved. This, of course, also increases power consumption, size, and
cost. For these cases, single chip processors are the preferred choice. Single chip processors can
be general purpose microprocessors (such as the Intel 80486), digital signal processors (such as
tasks (such as the Harris HSP50016 Digital Downconverter). Some specialized radio receiver
applications may not be bound by stringent physical size and cost limitations. Therefore,
processors of all types are considered in this report, ranging from single chip general purpose
Computations in digital signal processing can be performed using fixed-point arithmetic or
The advantage of floating-point arithmetic over fixed-point arithmetic is that it permits the use of
numbers with a much greater dynamic range. This is important in many digital signal-processing
27
In fixed-point arithmetic, the position of the decimal point in the register where each operand is
stored always is assumed to be the same. In floating-point arithmetic, each operand is
represented by a number stored in a register representing a fraction or integer. A number stored
in a second register specifies the position of the decimal point of the number stored in the first
register. Some processors do not have floating-point hardware and require floating-point
operation to be implemented in software. Software implementation of floating-point arithmetic is
typically much slower than hardware implementation.
Because floating-point operations are so important in digital signal processing, the speed of
processors is often specified in terms of millions of floating-point operations per second
(MFLOPS). This parameter allows comparison of the processing speed of different processors
and also allows determination of the time required to execute certain algorithms.
Many different benchmarks (such as the SPEC benchmarks, Whetstone, Dhrystone, and
Linpack) are used to compare speeds between processors. Each benchmark provides a number
indicating the relative speed of processing based on testing varying tasks. Results from the
application of a benchmark to different processors can be compared. However, results between
different benchmarks, in general, should not be compared. While these benchmarks are useful for
comparing processor performance, the parameter chosen to compare processing speeds between
processors in this report is the theoretical peak MFLOPS. This parameter was chosen due to its
ease of availability for virtually all floating-point processors, its lack of dependence on specific
benchmarking algorithms, and its relevancy to dedicated applications used in implementation of
a radio receiver. The theoretical peak MFLOPS parameter gives the maximum possible speed of
performance for the processor. It is found by computing the number of floating-point additions
and multiplications (using the processor’s full precision) that can be performed during a given
time interval [25].
Some examples of the processing speed of various types of processors ranging from single chip
processors to supercomputers are presented in Table 3. This table only gives a sampling of the
range of capabilities that exist in digital signal processing. Many other processors, with varying
capabilities, either exist or have been proposed. In addition, new developments with increasing
capabilities are announced all the time. An extensive listing of processing speeds for many
different computers is found in [25].
In certain situations, especially in the high-throughput case, overall processing performance is
not limited by the processor speed but by the maximum data transfer rates of the peripheral
components such as memory or I/O (input/output) ports. The inclusion of these factors in
platform evaluation should not be ignored when choosing a processor.
Table 3. Examples of Processing Technology
Processing
Speed*
Number of
Processors
Platform
Manufacturer and Model
50 MFLOPS
1
DSP Chip
Texas Instruments TMS320C40
28
Processing
Speed*
Number of
Processors
Platform
Manufacturer and Model
120 MFLOPS
1
DSP Chip
Analog Devices ADSP-21060/62
400 MFLOPS
8
VME Board
Pentek 4285
800 MFLOPS
16
Computer
Workstation
SUN Sparc 2000
6.48 GFLOPS
4
Supercomputer
Convex C4/XA-4
32 GFLOPS
4
Supercomputer
Hitachi S-3800/480
184 GFLOPS
3680
Massively
Parallel Computer
Intel Paragon XPS140
236 GFLOPS
140
Massively
Parallel Computer
National Aerospace Laboratory
Numerical Wind Tunnel (Japan)
* Theoretical peak processing speed.
Chia sẻ với bạn bè của bạn: