560 likes | 670 Views
Digital Fields Board (DFB) FPGA. Ken Stevens University of Colorado - LASP. Presentation Overview. Design Context Requirements Changes/Trades since PDR Module-level Design Examples FPGA Verification and Validation Strategy Current Status. DFB Block Diagram.
E N D
Digital Fields Board (DFB)FPGA Ken Stevens University of Colorado - LASP
Presentation Overview Design Context Requirements Changes/Trades since PDR Module-level Design Examples FPGA Verification and Validation Strategy Current Status
FPGA Top-level: Block Diagram Backplane Interface Analog Muxes and ADCs SRAM #1 SRAM #2
FPGA Top-level: THEMIS Heritage Backplane Interface Analog Muxes and ADCs SRAM #1 SRAM #2
FPGA Top-level: Changes Since PDR Backplane Interface Analog Muxes and ADCs SRAM #1 SRAM #2
FPGA Top-level: Implementation Status Backplane Interface Analog Muxes and ADCs SRAM #1 SRAM #2
FPGA Functional Overview (1/3) • The DFB FPGA accepts commands and provides telemetry to the DCB over a digital backplane: • Command Controller: • Decodes backplane protocol: ‘1’ start bit, 8-bit command ID, 16-bit command data field, odd parity bit, ‘0’ stop bit • Performs command translation to create appropriate register read/writes from/to sub-modules. • Data Controller: • Aligns telemetry data to increments of 1/128 second aligned to the 1 pps signal • Assigns APIDs • Encodes ALU data products for backplane transport to the DCB using a synchronous UART interface: ‘1’ start bit, 8-bit APID, 16-bit data field, odd parity bit, ‘0’ stop bit
FPGA Functional Overview (2/3) • The DFB FPGA controls acquisition of the analog signals and converts them to a database of signals at various rates for further processing: • Analog to Digital Conversion • 16-bit resolution • 16.384 kS/s (214) sample rate • Cross-strapped ADCs for limited redundancy • Field Alignment • Rotates E and SCM values into the coordinate system of the DC magnetic field • Digital Filtering • Multi-tap FIR low-pass filter that creates output rates from the sampling rate down to 1/(214)x sampling rate…i.e. [16.384kS/s, 8.192kS/s, … 2S/s, 1S/s] • Band-pass filter taps available for trigger value calculations
FPGA Functional Overview (3/3) The DFB FPGA generates the following types of data in order to meet science requirements (configured by the DCB): Trigger Data: pseudo-power in 7 or 13 frequency bands in the range of 0 to 8kHz Any two EDC, EAC, or SCM values Time Series (Waveform) Data: Reporting rates from 1s/s – 16Ks/s: V1-V6, V1AC-V6AC, E12DC-E56DC, E12AC-E56AC, FM1-FM3, SCM1-SCM3 Spectra Data: 2048-point FFT with selectable frequency binning and time averaging Any eight signals except V1-V6 Cross Spectral Data: Selectable frequency binning and time averaging Any four pairs from spectra data Solitary Wave Detector Data: Counts peaks and bins them according to magnitude E12AC or E34AC
Implementation Approach General Approach Using an FPGA-based implementation allows hardware to be tailored to the mission requirements Lower power than an equivalent processor based DSP design Heritage from THEMIS LTC1604 ADC and ADC support circuitry Digital Signal Processing algorithms Backplane interface protocols Changes and Improvements from THEMIS Higher sample rates (16ks/s) Addition of onboard Cross Spectral Analysis and Solitary Wave Detection Use of a single RTAX2000 FPGA instead of three SX72 FPGAs Rewrite of FPGA VHDL code Higher performance implementation of DSP algorithms Utilization of RTAX internal RAM to improve data flow and performance Parameterization of all modules to accommodate unforeseen design changes Standardization of all internal interconnect buses to Wishbone architecture
Trades & Changes Since PDR Added circuitry and an additional external SRAM to provide synchronization of data transfer to the 1 pps signal… Why? …to align data to increments of 1/128 second (the period of the buffer swapping in the DCB) coincident with the 1 pps signal so that DCB can correctly buffer and time-tag the data. How? …added double buffers in the data controller that are capable of holding 1/128 of a second of data. Data is tagged at acquisition and placed in the appropriate FIFO depending on when the sample was taken. Latency through the DSP is less than 1/128 second for all data types. Data is then burst onto the backplane at the next 1/128 second marker. Currently implemented and is in use in the lab using internal FPGA RAM but cannot hold the full data rate of 4096 words per 1/128 second period. An additional external SRAM has been added to the next board layout to facilitate 4096 word buffers for full rate data.
Module-level Design Examples Triggers and Waveforms are included only as backup slides since they were presented at PDR. Spectra/X-spectra illustrated to give an idea of the steps in the design process. Solitary Wave Detector to illustrate a “single page” overview.
Spectra/X-Spectra Data Path Backplane Interface Analog Muxes and ADCs SRAM SRAM
Spectra/X-Spectra Functional Diagram FFT Generation: - Select Input Signal and create 2048-point buffer - Hanning Window - Perform 2048-point FFT using Radix-2 Algorithm - End up with 1024 Real and 1024 Imag points Spectra Power: - Calculate power by adding squares of Real and Imag - Perform frequency binning - Accumulate in time - Pseudo log compress - Report at selected cadence X-Spectra: - Calculate X-Spectra components (see diagram) - Perform frequency binning - Accumulate in time - Pseudo log compress - Report at selected cadence
Spectra/X-Spectra Design Partitioning Control registers: Implemented in the top-level module Uses Wishbone interface Ingress Processor Compares incoming data from the digital filters with the selection registers and creates a 2048 sample buffer in external RAM for each enabled quantity Notifies the Control State Machine when input buffers are available Control State Machine (CSM) When an input buffer is available, the CSM transfers the data from the external RAM to an internal RAM, manages the calculation of a spectra, and then transfers the results back into external RAM. When the spectra calculations are complete, the CSM checks to see if any X-Spectra are enabled. If so, then the appropriate spectra are transferred to internal RAM, the CSM manages the calculation of the X-Spectra, and then transfers the results back to external RAM.
Spectra/X-Spectra Design Partitioning Arithmetic Logic Unit Implements the arithmetic logic required to do Windowing, FFT, Unpacking, Power Calculation, and X-Spectral analysis…will show data flow for each of these operations shortly Performs frequency binning Performs time averaging Egress Processor Transfers spectra and x-spectra results from external RAM Performs 34 to 8 pseudo-log compression Transmits data to the telemetry stream at the appropriate cadence.
Spectra/X-Spectra ALU Spectra/X-Spectra ALU is 32-bit fixed-point with appropriate rounding.
Spectra/X-Spectra ALU: Hanning Window …other operations are included in the back-up slides.
Solitary Wave Detector Data Path Backplane Interface Analog Muxes and ADCs SRAM SRAM
Solitary Wave Detector • Detect and Count Solitary Waves • Algorithm Overview: • Collect data samples in a buffer of 32, 64,128, or 256 samples • Calculate • The average is calculated and subtracted from each sample in the buffer to normalize the data. • If the highest absolute value in the buffer is more than 4X greater than pRMS, then count the peak as a solitary wave event • Events are counted in one of four counters based on software programmable threshold values • Reporting period is selectable from 0.5s to 32s in 0.5s increments • At the end of the reporting period, the four count values are pseudo-log compressed from 12 bits to 8 bits and sent to the TM module
FPGA Verification Approach Verification is performed independent of design Verification performed at both the module & chip levels Functionality and algorithm correctness is verified through module-level simulation Interaction with other parts of the system is verified through chip-level simulation Assertion Based Verification (ABV) Using Active HDL Code Coverage used to quantify verification Using Active HDL
FPGA Validation Plan Signal into DFB Analog Electronics From ADC to FPGA FPGA processing FPGA output Compare FPGA, Simulation & IDL results Record (raw) waveforms at 16kHz IDL routines performing same tasks as FPGA IDL formatting of data for ALU module simulation FPGA Simulation
Configuration Validation Plan • Verify commanding functionality • Start by testing nominal / default configuration • Can not test every possible configuration • (thousands for filter banks alone) • Instead, test each configuration option fully, one at a time (31 possible for filter banks)
Configuration Validation Plan S1(10) S2(10) Speed(11) En (2) En(2) Bands(2) E12 DC E12 DC 1/16 S/s On On 7 bins E34 DC E34 DC 1/8 S/s Off Off 13 bins E56 DC E56 DC 1/4 S/s … … … SCM3 SCM3 64 S/s
Configuration Validation Plan E12 DC 7 bins On On 1/8 S/s E34 DC S1(10) S2(10) Speed(11) En (2) En(2) Bands(2) E12 DC 1/16 S/s E34 DC Off Off 13 bins E56 DC E56 DC 1/4 S/s … … … SCM3 SCM3 64 S/s
Configuration Validation Plan E12 DC E34 DC E56 DC SCM3 7 bins On E34 DC On 1/8 S/s S1(10) S2(10) Speed(11) En (2) En(2) Bands(2) E12 DC 1/16 S/s Off Off 13 bins E56 DC 1/4 S/s … … … SCM3 64 S/s
DFB FPGA: Design Process ETU 1&2: Used to verify algorithms and functionality Reprogrammable FPGAs – ProASIC3e 1xSRAM to implement either FFT or 1 PPS in external RAM Currently in use at LASP and Berkeley ETU 3: Used to verify flight board layout and AX/RTAX specific logic Add EDAC RAMs where appropriate AX2000 in a BGA package with adapter to flight pin pattern 2x SRAM to implement both FFT and 1 PPS in external RAM FPGA to verify flight board characteristics at end of October FM1: Implements flight functionality Pinout change because of BGA to QFP adapter used on ETU 3 Flight functionality at end of 2009
Key Implementation Guidelines • Specification • Digital Fields Board Specification (RBSP_EFW_DFB_001D_SPEC Rev D) • Digital Fields Board FPGA Specification – In Progress • Coding Standards • LASP FPGA Design Guide • NASA 500-PG-8700.2.8 • Actel HDL coding Style Guide • Single clock domain • Fully synchronous design • Power-on reset is asynchronously applied, synchronously cleared • Scripts used for running simulations and builds to ensure consistency • Version Control • All source code, design documents, and test files are version controlled using Subversion • Major releases are archived in Agile CM system
Current Status • Peer review completed 9/10/09. Actions and responses are included in backup slides. • Working on algorithmic verification of spectra/x-spectra and solitary wave detector. • Using waveform and trigger functions in the lab to validate board layout, etc. • Expect ETU3 functional at end of October, FM1 at end of 2009
DFB FPGA: Field Alignment Flow Diagram E12,34,56 B (from FGM) Adjust Gains, Offsets Offsets 5-minute lowpass filter to find offsets Adjust gain/offset Rotation matrix Gains Offsets (from ground) Rotate to E system (gains included) Adjust gain/offset Interpolate Rotate E, B simultaneously in the xy plane until Bx = 0. Perform rotations using CORDIC Rotate E, B simultaneously in the xz plane until Bz = 0. E_perp = Ey/1.646 E_par = Ex/1.6462
DFB FPGA: Digital Filters • Uses a cascading filter algorithm that allows generation of both low-pass and band-pass outputs. • Each filter bank requires about the same processing as 2 comparable filters (e.g. 5th order Bessel). • Since quantities are typically filtered to so many different rates, and since some quantities require a filter bank anyway, this approach saves complexity and power. • -3dB point = 0.6 fNyquist • Power at fNyquist= -12 dB • Rolloff ~ 25 dB/octave • No phase shift
DFB FPGA: Triggers • Selects appropriate band-pass filter value from the digital filters output • The power in a particular band is approximated by averaging the absolute value • The results are used mainly in triggering algorithms
DFB FPGA: Digital Filters Diagram Low pass section Bandpass section 8 kS/s + - Shift Z-3 7-tap FIR filter 2-4 kHz Averager 4 kS/s 2:1 Decimating FIR filter (3-tap) + - Shift Z-3 7-tap FIR filter 1-2 kHz Averager 2 kS/s 2:1 Decimating FIR filter (3-tap) (9 more banks) (9 more banks) + - Shift Z-3 7-tap FIR filter 1-2 Hz Averager 2 S/s 2:1 Decimating FIR filter (3-tap)