Digital Signal Processors principles, use & application to PS systems.

Transcription

1 Digital Signal Processors principles, use & application to PS systems. Maria Elena Angoletta PS Seminar, 30 May 2002

2 TOPICS 1. Overview & history 2. Current scenery 3. Features 4. DSP choice criteria 5. Development environment 6. Current applications at PS - outline 7. Evolution at PS - hints 8. Conclusions M. E. Angoletta - DSPs principles, use & application to PS systems 2 / 25

3 Overview What is a Digital Signal Processor? A µ-processor for DSP-ing applications. Fast data processing. Predictable behavior. Re-programmable. DSPs commonly found in: Comms. (mobile handsets, wireless devices / Internet, GPS). Multimedia (printing, audio/speech/video/image processing). Control & data acquisition (biomedical, motor/process control). M. E. Angoletta - DSPs principles, use & application to PS systems 3 / 25

4 Overview CERN Overview CERN CERN-made DSP boards - Cost cutting. - Specific application needs. Ex: power supplies, experiments. COTS boards - Fewer modules needed. - Emphasis on signal processing. Ex: control, diagnostic applications. M. E. Angoletta - DSPs principles, use & application to PS systems 4 / 25

5 History Fixed point DMA, Floating point TI ~ 70% market. AD also popular. VLIW & SIMD architectures Development Consolidation MPD7720 TMS320C10 AT&TDSP32 TMS320C30 TMS320C6xxx BLACKFIN ADSP2100 TMS320C5xxx TigerSHARC TMS320C40 DSP56000 M. E. Angoletta - DSPs principles, use & application to PS systems 5 / 25

6 Current scenery Current scenery Performance Low : ~ 25 to 50 MHz clock, low cost / power consumption. Mid : ~ 150 MHz clock, multiprocessing support. High : Enhanced architectures - VLIW (Very Long Instruction Word) or SIMD (Single Input Multiple Data). Trend Blurred borderline to µ-processors, µ-controllers. FPGAs / SOPCs as alternative / help. If / which DSP? Global cost & performances difficult choice. Ex: Help from PS DSP Advisory committee. M. E. Angoletta - DSPs principles, use & application to PS systems 6 / 25

7 Features DSP molded by DSPing algorithms. Example y = N 1 i= 0 a i x i Fast memory access / data transfer. Fast computation (data path). Others FIR - Dedicated address generation unit. - Specialized instruction sets. M. E. Angoletta - DSPs principles, use & application to PS systems 7 / 25

8 Features - memory Features - memory Achieving fast I/O & memory access. 1. DSP architecture: (modified) Harvard. 2. Multiple-access memory: fast memory, multiple, sequential access per instruction cycle. 3. Multiported memory: multiple, independent, parallel access. 4. Memory access: reduced by program cache (zero overhead looping). 5. External memory interface: DMA (Direct Memory Access), memory data transfer without DSP involvement. M. E. Angoletta - DSPs principles, use & application to PS systems 8 / 25

9 Features - memory / 2 Features - memory / 2 DSP architecture PROGRAM MEMORY PM Address bus CPU DM Address bus DATA MEMORY Instructions & Data PM Data bus Instruction Cache DM Data bus Data Modified Harvard s: separate memories for data & instructions/data, instructions cache. improved memory access bandwidth. M. E. Angoletta - DSPs principles, use & application to PS systems 9 / 25

10 Features - computation Features - computation Registers: intermediate & final results, counters. Multiplier: one instruction cycle multiplication (& accumulation). Arithmetic Logic Unit (ALU): one instruction cycle ops. (basic arithmetic & logical). Shifter (simple / barrel): shifts (scaling) & rotates. M. E. Angoletta - DSPs principles, use & application to PS systems 10 / 25

11 Criteria for DSP choice Criteria for DSP choice 1. Arithmetic format (fixed vs. floating point, data width). 2. Performance. 3. Others (External interfaces, multiprocessing, cache memory, cost, development ease, power consumption.) A difficult choice! - Fragmented architecture & specialised features. - Objective & unified performance metrics set lacking. M. E. Angoletta - DSPs principles, use & application to PS systems 11 / 25

12 Criteria: arithmetic format Criteria: arithmetic format DSP Fixed-point Floating-point 16-bits 32-bits 20-bits 24-bits IEEE bits Others Native data word width Data width manipulated by buses & data path in single instruction cycle. The wider, the costlier! M. E. Angoletta - DSPs principles, use & application to PS systems 12 / 25

13 Criteria: arithmetic format Criteria: arithmetic format Parameters Dynamic range db = log largest smallest value value High dynamic range wide data set representation without overflow. NB: Different applications have different needs. Ex: telecomms = 50 db, high fidelity audio = 90 db. Max. precision bits = log 2 max value max quantisation error M. E. Angoletta - DSPs principles, use & application to PS systems 13 / 25

14 Criteria: arithmetic format Criteria: arithmetic format Fixed & floating : advantages Fixed-point Simpler h/w Less Silicon - Lower cost. - Higher CPU speed. - Less power consumption. Floating-point High dynamic range (32 bits: ~1500 db float vs. ~180 db fixed). Reduced coding effort (no scaling needed). More precision ( BUT word length dependent). M. E. Angoletta - DSPs principles, use & application to PS systems 14 / 25

15 Criteria: arithmetic format Criteria: arithmetic format Arithmetic format choice - summary 1. ADC / DAC bits. Floating-point above 12 / 14 bits. CAVEAT : noise floor modulation. Ex.: HiFi audio industry (20 to 24 bits) often prefers fixed-point. 2. Algorithm complexity / changes. Floating-point for complex & upgradable algorithms. 3. Product & development costs. Choice based on number of units produced. Ex.: expensive development affordable by high-volume applications. M. E. Angoletta - DSPs principles, use & application to PS systems 15 / 25

16 Criteria: performance Criteria: performance Performance commonly judged via metrics set. Metrics set - Max clock frequency [MHz] - Power consumption [W or W/MIPS] - Execution throughput [MIPS, MOPS] - Memory bandwidth [MB/sec] latency [clock cycles] - I/O Ports No. Often misleading! - Mostly peak values. - Fuzzy interpretation. Alternative: use benchmark data (kernel / application). M. E. Angoletta - DSPs principles, use & application to PS systems 16 / 25

17 Development environment Development environment Debugging In-chip emulator - Pod-based. - Scan-based (JTAG, OnCE). Code development Standard Integrated Development Environment (IDE) Often vendor-provided (ex: TI s CodeComposer) Languages: assembler, C, C++ (also ADA). RTOS (sometimes). Advanced Matlab DSP rapid prototyping suite RIDE (Hyperception) visual real-time design. M. E. Angoletta - DSPs principles, use & application to PS systems 17 / 25

18 Development env ment /2 Development env ment /2 Example of Code Composer IDE. M. E. Angoletta - DSPs principles, use & application to PS systems 18 / 25

19 Current applications at PS Current applications at PS Outline Reliable operational systems: 3 BD + several PO. COTS boards in BD - beam parameters measurements ( tune / intensity / p) - machines: PS (1994), PSB (1998), AD (2000). - DSPing tasks. Custom boards in PO. - DSP for digital regulation, - CPU for supervision & Control Room interface. CAVEAT - out-of-market h/w, require proactive action, - systems consolidation (maintenance). M. E. Angoletta - DSPs principles, use & application to PS systems 19 / 25

20 AD multi-diagnostics (1) AD multi-diagnostics (1) COTS DRX Pentek 6510 C40 floating-point DSP. Noise generator PU Longitudinal PU H Transversal PU V Transversal DRX-based, beam multidiagnostic system. System measures - intensity, - momentum, - momentum spread, ADC Controls & setup - tune H & V (BTF method). PS/BD Talk 18/06/02 DRX board Pentek 6510 VMEbus DSP TMS320C x HSP50016 DDC POWER PC (1) M. E. Angoletta, V. Chohan, M. Ludwig, O. Marqversen, F. Pedersen, The new digital-receiver-based system for antiproton beam diagnostics, PAC 01 M. E. Angoletta - DSPs principles, use & application to PS systems 20 / 25

21 PS tune system (2) PS tune system (2) COTS VASP-16. C25 fixed-point DSP. 4 Zoran Vector Signal Processors (VSP). Board / DSP out of production 1 board for operation & 1 spare. System measures tune H & V. Controls MVME147 PU H Transversal MEM B TMS320C25 DSP MUX PU V Transversal ADC VASP16 MEM A 4xZR34161 VSP VMEbus (2) J. Gonzalez, S. Johnston, E. Schulte, Fast Q-measurement for the PS by FFT analysis, EPAC 94 M. E. Angoletta - DSPs principles, use & application to PS systems 21 / 25

22 PSB tune system (3) PSB tune system (3) PU H Transversal PU V Transversal COTS Board DataBeta DBV96. Motorola floating-point DSP. Board / DSP out of production 1 board for operation & 7 spares. System measures tune H & V. Controls ADC MEMORY CARD ADC DBV96 POWER PC DSP96002 VMEbus (3) A. Chapman-Hatchett, V. Chohan, E. T. d Amico, Tune measurement for the CERN Proton Synchrotron Booster Rings using DSP in VME, PAC 99 M. E. Angoletta - DSPs principles, use & application to PS systems 22 / 25

23 Power supply regulation (4) Power supply regulation (4) Custom boards (~ 140). DSP: regulation Input power POWER - Loop = 100 µs. - Motorola / fixed-point DSPs. - PID / test modes. CPU: supervision, CR interface & local control. Fault detection: protection system. FAULT DETECTION HW To / from Control Room Internal bus CPU MOTOROLA DSP Local display / keyboard GFAS DAC Logic ADC RACK (4) Source: J-L Gomez-Costa. Development by LIC Argentina. M. E. Angoletta - DSPs principles, use & application to PS systems 23 / 25

24 Evolution at PS Evolution at PS HINTS 1. Consolidation of tune measurement systems. 2. LEIR (Low Energy Ion Ring) tune measurement. 3. Linac RF new low-level control system. 4. Status to be re-evaluated in M. E. Angoletta - DSPs principles, use & application to PS systems 24 / 25

25 Conclusions 1. DSP is a hot topic - lot of interest at CERN. 2. Powerful & well developed - no fancy! 3. Several systems operational at PS. 4. Foresee maintenance, upgrade & improvement. 5. Initiatives to encourage DSP usage. Ex: DSPing / DSP courses. 6. Varied choice (DSP, FPGA etc) - appropriate selection difficult. Help from PS DSP Advisors. 7. More info: DSP Training Course ~ early M. E. Angoletta - DSPs principles, use & application to PS systems 25 / 25