Slide 1: DSP Processor Architectures Presented by
Saroj Kumar Panda
2003EEZ0019 Slide 2: • Digital audio applications
– MPEG Audio
– Portable audio
• Digital cameras
• Cellular telephones
• Wearable medical appliances
• Storage products:
– disk drive servo control • Industrial control
– Base station
– Cable modems
• Military applications:
– sonar DSP Applications Slide 3: TYPES OF DSP PROCESSORS
• 32-BIT FLOATING POINT (5% of market):
– TI TMS320C3X, TMS320C67xx
– AT&T DSP32C
– ANALOG DEVICES ADSP21xxx
– Hitachi SH-4
• 16-BIT FIXED POINT (95% of market):
– TI TMS320C2X, TMS320C62xx
– Infineon TC1xxx (TriCore1)
– MOTOROLA DSP568xx, MSC810x
– ANALOG DEVICES ADSP21xx
– Agere Systems DSP16xxx, Starpro2000
– LSI Logic LSI140x (ZPS400)
– Hitachi SH3-DSP
– StarCore SC110, SC140 Slide 4: • Data path configured for DSP
– Fixed-point arithmetic
– MAC- Multiply-accumulate
• Multiple memory banks and buses -
– Harvard Architecture
– Multiple data memories
• Specialized addressing modes
– Bit-reversed addressing
– Circular buffers
• Specialized instruction set and execution control
– Support for fast MAC
– Fast Interrupt Handling
• Specialized peripherals for DSP Basic Architectural Features of DSPs Slide 5: • DSP processors are microprocessors designed for efficient mathematical manipulation of digital signals.
• DSP in-sensitive to environment
• DSP performance identical even with variations in components;
analog systems behavior varies even if built with same components with 1% variation Basic Features of DSPs Slide 6: • DSPs tend to run one program, not many programs.
– Hence OSes are much simpler, there is no virtual memory or protection, ...
• DSPs usually run applications with hard real-time constraints:
• DSPs usually process infinite continuous data streams.
• The design of DSP architectures and ISAs driven by the requirements of DSP algorithms. Basic Features of DSPs Slide 7: • The algorithms of DSPs :
– Inifinite Impule Response (IIR) filters
– Finite Impule Response (FIR) filters
– FFT, and
• In DSPs, target algorithms are important:
– Binary compatibility not a mojor issue
• High-level Software is not (yet) very important in DSPs. Basic Features of DSPs Slide 8: Architecture of the Digital Signal Processor Slide 9: Architecture of the Digital Signal Processor Slide 10: Programmable and Application-specific DSP processors Systems with programmable processors are called multiprocessor system Systems with application-specific processors are called array system Array processors should have the following characteristics
Regularity and modularity
Spatial and temporal locality
Parallelism and Pipelinability Slide 11: Design Process for Array Processors Programmable Digital Signal Processors Slide 12: Systolic Architecture 100ns 5 MOPS AT MOST
The conventional processor 100ns 30 MOPS AT MOST
A systolic processor array Programmable Digital Signal Processors Slide 13: Programmable Digital Signal Processors Systolic Architecture Characteristics
Each data item can be used and re-used before the final result reaches the memory.
Several data items can be processed concurrently.
The operations performed by the processors at each step are simple.
The architecture has a regular geometry, and the flow of control and data is regular and consistent across the processors.
The regular and localized communication geometry reduces routing costs, power, time and chip area required to implement a computation Slide 14: CORDIC Architecture The basic of the algorithm is the coordinate rotation in a linear, circular or hyperbolic coordinate system depending on which function is to be calculated iteratively with simple shift and add operations
By varying a few simple parameters, the same CORDIC processor is capable of iteratively evaluating this elementary function using the same hardware within same amount of time.
The CORDIC algorithm can also be implemented with VLSI to achieve high throughput computation in real time signal processing Slide 15: Data Path Synthesis This is used to designate the transformation of a behavioral description to a structural representation.
Its aim is to determine a suitable hardware structure as automatically as possible from a behavioral description.
Data path synthesis allows systematic derivation of the structure of a data path to be synthesized. Programmable Digital Signal Processors Slide 16: Starting from a CDFG, the data path synthesis comprises three major subtasks: Scheduling :
Assigning operations to time or control steps, taking into consideration the dependences between operations (2) Allocation :
Selection of functional and storage units (3) Binding :
Binding of units Programmable Digital Signal Processors Data Path Synthesis Slide 17: Programmable Digital Signal Processors Scheduling and allocation for simple CDFG S3 = X0 x W0 + X1 x W1 + X2 x W2 Slide 18: Programmable Digital Signal Processors DIF Array Processors for one-dimensional DHT Slide 19: Programmable Digital Signal Processors DIF Array Processors for one-dimensional DHT Slide 20: DIT Array Processors for one-dimensional DHT Programmable Digital Signal Processors Slide 22: References: Architectures for Digital Signal Processing by Peter Pirsch [WILEY]
The Scientist and Engineer’s Guide to Digital Signal Processing [ www.ti.com]
W.H. Fang and J.D.Lee,”Efficent CORDIC-based systolic architectures for the DHT”,IEEE Proc., Comput. Digit. Tect., Vol.142,pp201-207, May 1995