gps/ins computing system

Company

LOGO

Final presentationSpring 2008/9

Performed by:Alexander Pavlov David Domb

Supervisor:Mony Orbach

GPS/INS Computing

System

GPS/INS Computing

System

AgendaAgenda

1. General overview1. General overview

2. Our Project2. Our Project

4. Results4. Results

GPS/INS Computing System

3. The Design3. The Design

5. Summary5. Summary


General overviewGeneral

overview“Even Noah got no salary for the first six months partly on account of the weather and partly because he was learning navigation.”

Mark Twain

Theoretical Navigation Algorithm Theoretical Navigation Algorithm

0

•Initialization

1

•Particle Propagation

2

•Particle Update & Normalization

3

•State Estimation

4

•Effective N calculation

5

•D computation

6

•Re-sampling

7

•Regularization

8

•Weight Re-computation


Developed in the “Technion” and Implements the tightly coupled INS/GPS navigation unit, with the particle filter.

The algorithm stages:

Project GoalsProject Goals

Establishing the efficiency of the particle filter based, tightly coupled INS/GPS navigation unit realization.

Designing an efficient real-time particle filter based, tightly coupled INS/GPS navigation unit.


GPS Computing System

Our Project

GeneralGeneral

• Our goal was to implement Particle Propagation and State Estimation stages.

• Both stages were required to function within 0.01 sec.


Group Project GoalsGroup Project Goals

Implementation of Particle Propagation and State Estimation stages of algorithm

Successful integration with other groups for evaluating the entire algorithm’s implementation.



TheDesign

Solution – Top designSolution – Top design


Weight vector

Particles propagation

unit

State estimation

unit

Estimated State Vector

[1..18]

xN Extended State Vector

[1..18]

Extended State Vector

[1..18]

Extended State Vector

[1..18]

Co

ntro

ller

Basic architectureBasic architecture

• 24Bit words data bus.• FIFO-Like streaming interfaces

( Request + Empty / Full )• Controlled By Start/Finished activation mechanism

BasicStreaming

Block

Start Finished

Data in

Write requestFull

Data outRead request

Empty

Control

Input Path

Output Path

Particle propagation unitParticle propagation unit


clockreset

start

finish

ParticlePropagation

Unit

X[0..439]

INS[0..287]

X_OUT[0..439]

Particle propagation unitParticle propagation unit


Propagation Unit 1

Propagation Unit 2

Propagation Unit 6

MUX(6 to 1)

Propagationtimingcontrol

Single particle propagation data flowSingle particle propagation data flow

Format inputs to 48 bits

Calculate trigonometric

functions• Latitude sin/cos

Format trigonometric

function output to 48 bits

R_E, R_e, R_N calculation

Denominator calculation

• d_longitude denominator• d_latitude denominator

Dividers• d_longitude• d_lattitude• R_e

Particle

Propagation


Propagationflow

control

Estimation unitEstimation unit


clockreset

New_Data_In

Estimation_Ready

EstimationUnit

X[0..439]

W[0..23]

ESTIMATED_DATA[0..439]

Estimation unitEstimation unit


W

X Σ Estimated Data×


RESULTS

Physical implementationPhysical implementation


Physical implementation of entire design was unsuccessful due to lack of FPGA resources.

Therefore, only 1 of the 6 parallel “propagation unit” blocks was implemented.

Resources utilizationResources utilization


Base design(Without our

Logic)

Base + Our design

(Without trig Logic)

Base + Full design

(Including trig Logic)

Resources analysisResources analysis


A design with 6 prop units will need approximately: • 130K combinational ALUTs (85K available).• 162K logic registers (85.2K available).• 20M block memory bits (8.25M available).• 4074 DSP blocks (896 available).

Possible FPGAs:• Xilinx – Virtex6 / 7.• Altera – Stratix 5 (possible).

Timing AnalysisTiming Analysis


The implemented design of 1 prop unit produced:• Particle LATENCY – 97 clock cycles (from

“start” to “finish”) @100MHz = 1uSec:



The implemented design of 1 prop unit produced:• Throughput of 38 clock cycles (from “finish” to

“finish”) @100MHz = 380nSec



The total time with the implemented design of 1 prop unit produced was 30,000 particles in 1,140,059 100MHz clocks = 11.4mSec.

Note that the clock frequency of 100MHz was changed from the original plan of 30MHz, due to working with only one prop unit.

Accuracy resultsAccuracy results


We have encountered many problems while trying to test our results:• The “Generic program” for 1 FPGA did not

work correctly – we were unable to control the inputs to the design.

• The “Generic program” for 4 FPGAs did not work as anticipated with the SW data files:o The SW data input files were arranged not

according to the “bits order” agreed upon.o The program’s data output files did not

reflect the output values from our design correctly.

Accuracy resultsAccuracy results


We have made a manual accuracy check for one particle, by comparing the result as viewed with the “signal tap” tool to the SW result.

For the tested particle, we got a location result which was different from the SW result by 0.0002%:

SW RESULT OUR RESUL

0.0910404 0.0910405814647675

Group’s goals achievement Group’s goals achievement


Implementation of our design:• PARTIAL - due to lack of FPGA

resources.

Design testing and integration:• PARTIAL - due to problems with the

testing environments and no cooperation from other design teams (which finished their project).

Our conclusionOur conclusion


In terms of possibility – it seems that it is possible to implement the “Propagation” and “Estimation” stages of the project, within the necessary timing requirements, on a better, more powerful FPGA (without changing the design)

For integration with other projects, it is important to have the project’s teams present. Otherwise, it can’t be done efficiently.

gps/ins computing system

Documents