Senior design final presentation master

Post on 21-Feb-2017

83 views 1 download

transcript

SENIOR DESIGN PROJECT

SPRING 2016

Christian Ladigoski

Advisor: Dr. David Weissman

DESIGN OF A FREQUENCY MODULATED – CONTINUOUS WAVE (FM-CW) PEDESTRIAN-RECOGNITION AUTOMOBILE

RADAR AND ALGORITHMS FOR USE IN FOG

OUTLINE OF WORK• General Automobile Radar System Parameters

• Antenna Design, Response and Analysis

• Phased Array Design, Response and Analysis

• Design of the FM-CW Waveform

• Signal Processing

• Improved Performance in Fog

• Constant False-Alarm Rate Detection

• Pedestrian Recognition and Beam-scanning

Figure 1. Concept Behind Pedestrian Recognition

Figure 2. The many radar systems that currently exist in modern cars

GENERAL RADAR SYSTEM PARAMETERS• Problem statement: Pedestrian detection is a problem that becomes complicated with the

presence of fog; on a clear day, there is typically no problem seeing up to 100 meters in front of your vehicle – however, visibility decreases a great amount when the intensity of fog rises.

• Operating Frequency: 77 Ghz Wavelength of 3.9mm/0.0039m

• Two options:

• 76-77 Ghz (Narrowband / Long Range) – This is our design choice

• 77-81 Ghz (Wideband / Short Range)

• Waveform: Linear Frequency-Modulated Continuous Sinusoidal Wave

• Phased Array Antenna (Two Uniform Rectangular Arrays)

• Maximum intended range: ~100m

• Antenna Azimuth Beamwidth: ~12 degrees

• Antenna Elevation Beamwidth: ~16 degrees

Figure 3. Generalized Block Diagram of a Bistatic FMCW System

Time

Frequency

FIG. PATCH ANTENNA & [20 X 36] ULA

Figure 4. Patch Microstrip Antenna Element with Substrate

Figure 5. Layout of a 20x36 Element Array

Substrate

Element

ANTENNA DESIGN (1)• Patch Microstrip

• Operating Frequency: 77 GHz

• Length = 1.098 mm

• Width = 1.353 mm

• Height = 6.000 mm

• Dielectric Permittivity = 2.1 (Teflon)

• Dielectric Loss Tangent = 0.002

Figure 6. Inset Feed Patch Antenna

ANTENNA DESIGN (2)• In order to ensure reasonable power transfer to the antenna and account for return loss, impedance

matching is key.

• Implementation: MICROSTRIP LINE FEED

• Characteristic antenna impedance is approximately 177 ohms

• To match a 50 ohm transmission line, the inset feed would be placed at approximately L/3.1 into the center of the patch, where L = length of the element.

• Microstrip Line Feeding versus Coaxial Line Feeding

• Both struggle with thicker substrates

• Both are limited by narrow bandwidth

• Coaxial Line Feeding requires higher order modes due to asymmetries, which fosters cross polarization

• Microstrip Line Feeding is easier to manufacturer and cheaper, is easier to match (changing inset feed point)

Figure 7. 3D Directivity Pattern of the 20 x36 Phased Array

PHASED ARRAY (1)• Array is a matrix of patch elements, N x M (rows by columns) elements

• Design: 20 x 36, 1 Transmit Uniform Rectangular Array (URA), 1 Receive URA

• This translates to an elevation beam-width of approximately 12 degrees and an azimuth beam-width of approximately 16 degrees

• Element Spacing of approximately half of a wavelength (0.00195 Meters) to avoid deleterious effects that allow signals from unwanted directions from being observed

• Directivity describes the directionality of the beam of the array (a smaller solid angle equates to a higher directivity). Peak directivity of approximately 32 dBi at 0 degrees Azimuth / 0 degrees Elevation

Figure 8. Azimuth Cut of Beam (Elevation Angle is 0 degrees)

Figure 9. Elevation Cut of Beam (Azimuth Angle is 0 degrees)

LFM-CW WAVEFORM• Sample Rate = 150 MHz (Rate of transmission of identical pulses)

• Sweep Bandwidth = 300 MHz

• This yields a range resolution of 0.5m (Range resolution determines the ability to separate/resolve individual targets)

• Sweep Time = 2 microseconds

• Estimated with consideration that the maximum intended range is 100 meters.

• Sweep type: Double Triangular

• This was chosen not only to resolve ambiguity between range and Doppler but also to aid in a common FM-CW radar issue: ghost targets

• The sweep slope (2B/Tr) will differ for the second triangular sweep.

• We decided to test the feasibility of the waveform in detecting a moving target relative to a moving source.

Figure 10. Principles of how FMCW Waveforms Work

Figure 11. Double Triangular Sweep waveform

Figure. Power Spectral Density of the return signal; The (narrowband) dechirped signal obtained from a target echo utilizating FMCW

Figure 12. Range-Speed Response Pattern of the same target.Range is approximately 30 meters and speed (relative to source) is approximately 6.5 m/s.

IMPROVED PERFORMANCE IN FOG (1)There are two things to consider that could disrupt the signal: atmospheric attenuation and weather conditions such as rain and fog.

This first graph shows atmospheric attenuation -- dependent on humidity, temperature, and frequency of the radar (77 GHZ).

Attenuation at 77GHZ (Appx 0 dB/km)

APPX 77GHZ

Figure 13. Atmospheric Attenuation vs Frequency

IMPROVED PERFORMANCE IN FOG (2)The second causality includes attenuation caused by weather conditions -- in this case, fog.

Data was analyzed for medium density fog (at 300m visibility) and heavy density fog (at 50m visibility).

There is a slight offset from attenuation that is considered when detecting pedestrians.

Medium Fog = + 0.15 dB/kmHeavy Fog = + 1.5 dB/km

APPX 77GHZ

Attenuation at 1.5 dB/km

77GHZ

Figure 14. Fog attenuation vs Frequency

Figure 15. Specific Attenuation Constant for Fog (“K”) with respect to Frequency

Figure 16. Path Loss Due to Fog vs. Range

Figure 17. SNR with FMCW Signal and Fog Attenuation

(We need a minimum SNR of 5 dB to detect at max range of 100 meters)

SIGNAL PROCESSING – PEDESTRIAN RECOGNITION (1)

• To detect pedestrians, it is important to realize that the magnitude of their power return is typically much lower than that of a car, truck, etc. Likewise, their Doppler shift is typically expected to be much larger (relative to a moving car) than that of a moving vehicle nearly matching the source speed.

• Targets will then need to be classified based on two big factors:

• The magnitude of the returned power – this should give consideration to the fact that a pedestrian’s radar cross section (RCS) fluctuates greatly, and is decreased with the presence of fog. The expected range of a pedestrian RCS is from -5 dBsm (roughly 0.37m2 ) to 0 dBsm (1m2).

• Doppler spectrum – Doppler filters will need to be used.

• Classification and detection should be done within the time it takes for a car (going at a certain speed) to stop.

CAR SPEED AND STOPPING DISTANCES• Stopping Distance For a Car Going…

• 80 MPH: 306 Feet or 93.3 Meters

• 60 MPH: 172 Feet or 52 Meters

• 40 MPH: 76.5 Feet or 23.3 Meters

• 20 MPH: 19.11 Feet or 5.8 Meters

• 10 MPH: 4.77 Feet or 1.45 Meters

SIGNAL PROCESSING – PEDESTRIAN RECOGNITION (2)

Figure 18. Power Transmitted vs. Radar Cross Section of Pedestrian (Target to 100 meters)

SIGNAL PROCESSING – PEDESTRIAN RECOGNITION (3)

Figure 19. Power Received vs. Radar Cross Section of Pedestrian (14.5W of Transmitted Power)

SIGNAL PROCESSING – ANTENNA BEAM-SCANNING (1)• For our purposes, it is important to get an accurate direction-of-arrival estimate. Beam-

scanning is a method that will allow the array beam to be scanned over a region of interest and look for a target.

• We utilized beam-scanning to see the response when two targets are realized in a region of interest.

Figure 20. Two target case - one at 40° Az/9° El and another at -12° Az/-2° El.

Figure 21. Closely spaced targets, one at 0° Az/8° El and another at -2° Az/ 9° El.

SIGNAL PROCESSING – ANTENNABEAM-SCANNING (2)• As shown in the previous figure, beam-scanning by conventional means fails to resolve

closely spaced targets – crucial for our purposes. Thus, we decided to utilize a Minimum Variance Distortionless Response (MVDR) method to beamscan.

Figure 22. Two target case - one at 40° Az/9° El and another at -12° Az/-2° El.

Figure 23. Closely spaced targets, one at 0° Az/8° El and another at -2° Az/ 9° El.

SIGNAL PROCESSING – CONSTANT FALSE ALARM RATE (CFAR) DETECTION (1)

• A threshold is the level at which we expect the power of a return to be before being considered a target of interest. Threshold must be set at a “comfortable level” – that is, there must be a balance between probability of detection of a true target and probability of false alarm

• Threshold level:

• High: Low probability of detection, many missed targets

• Low: High probability of detection, many false alarms (Preferred)

• Must be made variable to account for interference, noise, cluttering, etc.

• Note:

• Further distances – influence of noise level is higher

• Closer distances – influence of clutter level is higher

• Therefore false alarm rate must take range into account

• (For our scenario, is it important to note that having higher probability of detecting noise, clutter and other interference above the threshold is preferred for safer driving conditions)

SIGNAL PROCESSING – C.F.A.R DETECTION (2)HOW DOES CFAR WORK?

• Method: Cell-Averaging CFAR Detector

• The circuit will estimate the noise/clutter level around the target cell to determine if the target cell is of our interest or not.

• Circuit will determine if the power threshold is considered to be above a level expected from a target of interest.

• We tested varying thresholds as to see which would result in a small probability of false alarms while still realizing pre-set targets at 35, 55, and 57 meters from the radar source.

Figure 24. Concept of CFAR Detection

Probably Detection at 0.9False Alarm Probability at 1e-6Targets at 35, 55 and 57 meters

Figure 25. Randomized Noise vs Range

Figure 26. CFAR Detection with High Threshold

Figure 27. CFAR Detection with Low Threshold

SIGNAL PROCESSING – CFAR DETECTION 3

Figure 28. Beam-scanning from -45° to +45° Azimuth (two targets)

Figure 29. Two targets meeting the threshold determined by noise power

Targets

ACKNOWLEDGEMENTS• Thank you to Dr. David Weissman, Professor Hausman as well as Hofstra University

• Gaussian Noise - http://www.gaussianwaves.com/2015/06/how-to-generate-awgn-noise-in-matlaboctave-without-using-in-built-awgn-function/

• CFAR - http://www.mathworks.com/help/phased/examples/constant-false-alarm-rate-cfar-detection.html

• Fog Detection - https://www.itu.int/dms_pubrec/itu-r/rec/p/R-REC-P.840-3-199910-S!!PDF-E.pdf

• http://www.mathworks.com/help/phased/ref/fogpl.html

• https://www.itu.int/rec/R-REC-P.840-6-201309-I/en

REFERENCES