The “Coffee-Can” Radar Redesigned as an Inexpensive RF PCB

■ Jim Carroll, Gent Paparisto, and David Vye

T
his article describes an inexpensive and small- 2.4-GHz industrial, scientific, and medical (ISM)-band
size radar RF core that can be integrated into radar employing connectorized RF components. In this
applications such as drones and unmanned article, the OCW work is extended by redesigning the
vehicles and opens the door to many projects requir- entire RF chain to reduce
ing better environmental sensing. Our work was costs, improve perfor-
inspired by the November mance, and provide an
2012 IEEE Spectrum RF core based on printed
article “Coffee-Can circuit board (PCB) tech-
Radar: How to Build nology. This enables the
a Synthetic Aper- use of inexpensive sur-
ture Imaging Sys- face-mount technology
tem with Tin Cans (SMT) components that
and AA Batteries” can be mass produced
[1], which describes and used more easily by
a simple radar system students and hobbyists.
that can be built with easily coffee can–©istockphoto.com/Sergio Bellotto,
circuit board— ©istockphoto.com/naqiewei
obtainable RF components. OCW Overview and
The Spectrum article recounts an engineer’s experi- Theory of Operation
ence constructing a radar system based on the Massa- The OCW system is based on a frequency-modulated
chusetts Institute of Technology (MIT) OpenCourseWare continuous-wave (FMCW) signal operating at the 2.4-GHz
(OCW) online material [2]. The OCW material details ISM band with about 10-mW transmit power and a maxi-
the operating theory, design, and construction of a mum range of approximately 1 km. The system is capa-
ble of estimating the range and velocity of a target
Jim Carroll (jim.carroll@ni.com) is director of IC design flows with the course’s baseband processing algorithms.
with AWR Group, National Instruments (NI), Allen, Texas, The basic system architecture and the functionality
United States. Gent Paparisto (gent.paparisto@ni.com) is in of the redesigned RF chain are the same as the origi-
RF systems product marketing at AWR Group, NI, Carlsbad nal system, enabling users to continue leveraging
California, United States. David Vye (david.vye@ni.com) is
the OCW material’s concepts, supporting circuitry,
director of technical marketing with AWR Group,
NI, Newburyport, Massachusetts, United States. and software processing tools with the redesigned
system. To facilitate this work, an electronic design
Digital Object Identifier 10.1109/MMM.2016.2589201
automation (EDA) software program was used for the
Date of publication: 8 September 2016 system analysis, RF circuit design, and layout of the

62 1527-3342/16©2016IEEE October 2016

RF PCB. The result was a radar core RF PCB designed
with full FMCW time-domain system simulations,
which allowed better-informed design and tradeoff
decisions to be made during the redesign.
The assembled MIT OCW radar is shown in F ­ igure  1,
including the coffee-can “cantennas,” core RF chain,
baseband breadboard circuitry, and battery packs
mounted onto an approximately 12-in × 15-in wooden
board. The device is a great learning tool because it
can be built with easily obtainable parts. However, the
total cost, size, and form factor of the device shown in
Figure 1 are disadvantageous for many applications
where users may want to integrate its radar-sensing
capability. Lowering the cost and shrinking the form
factor will enable the radar to be used in many more
applications than the original OCW radar.
For example, miniaturization of the radar onto a
small RF PCB makes the new system small and light
enough to be attached onto platforms such as an
autonomous drone. Also, creating a radar RF core that Figure 1. The assembled MIT OCW “coffee-can” radar [2].
is much less expensive will enable entire classrooms,
even at the high school level, to obtain test boards for
learning and experimentation. All of these goals can
be achieved by moving the OCW system to its next f (t)
natural development level. Transmit Receive
f max

FMCW Fundamentals
f min
The theory of operation for such systems is a good τ T–τ
T
place to start in understanding the basics of this radar t
redesign. A traditional pulsed radar detects the range f (t)
to a target by emitting a short pulse and observing the f bd
time of flight of the returned target echo. This requires f bu
t
the radar to have high instantaneous transmit power
and often results in a radar with a large, expensive
Figure 2. The frequency of the transmitted and shifted
physical apparatus.
received frequency-modulated signals and beat frequencies
FMCW radars achieve similar results using much as a function of time for a moving target.
smaller instantaneous transmit power and having a
smaller physical size by emitting a continuous micro-
wave signal that is frequency modulated with a low- mately sinusoidal with a frequency fW, which is con-
frequency waveform [3], the duration of which is much stant in the time interval T - x and equals the change
greater than the return time of the echo, such as a saw- of the transmitter frequency during time x,
tooth function of period T illustrated in Figure 2.
To simplify the analysis here, the echo signal fW = ax, (2)
received after reflection from a stationary object is a
copy of the transmitted signal, delayed by the propa- where a = Tf/T is a modulation waveform slope and
gation time: Tf = fmax - fmin is the maximum frequency deviation.
For a moving target, the received signal will also
x=
2R , (1) contain a Doppler shift term. A time-variant beat fre-
c
quency (which is the difference between the transmit-
where R is the target range and c is the speed of light. ted signal and Doppler-shifted received signal) will be
The transmitted signal is reflected off the target and generated for the up chirp and down chirp, denoted
received by the antenna and is then mixed with the fbu and fbd , respectively, as illustrated in Figure 2. The
signal from the transmitter. This type of RF chain is range beat frequency fr and the Doppler frequency fd
referred to as a homodyne system because it uses the are mathematically expressed as fr = ; fbu + fbd ; /2 and
transmit signal for down conversion. After low-pass fd = ; fbu = fbd ;/2.
filtering, a differential frequency [termed video or The range beat frequency fr and Doppler frequency
intermediate frequency (IF)] signal is obtained, approxi- fd can be obtained by signal processing of the video

October 2016 63
 is used for the down conversions of the received sig-
Mini-Circuit VCO Voltage and nal in this homodyne system, this deviation in the VCO
Output Frequency frequency will not significantly affect or bias the total

Frequency (MHz) Voltage (V)

5
Ramp 4 VCO Control Voltage system metrics.
VCO Output Frequency (MHz)
3
2
1 Original System Design
0 The original OCW system diagram’s center frequency
2,600
2,550 operates in the ISM band, which enables the use of
2,500
VCO

2,450 802.11bgn Band 14 widely available low-frequency components. Submin-

2,400 802.11bgn Band 1 iature version A (SMA) connectorized Mini-Circuits
2,350
2,300 RF components were used for the entire radar RF
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 core. Figure 4 illustrates the OCW system architec-
ture, in which the SMA components were directly
Figure 3. The OCW simulated voltage ramp and Mini- fastened together to connect each function. The base-
Circuits VCO output frequency.
band signal (IF) from the radar unit was amplified
and filtered using breadboard circuitry and standard
low-frequency operational amplifier components. A
laptop was then used to sample the resulting down-
Tx
PA converted IF spectrum using the computer’s sound-
Pad card. MATLAB scripts were used to perform the
VCO 50 Ω baseband signal processing, thereby extracting out the
distance and velocity estimations from the sampled
Rx
echo recordings.
IF LNA Figure 4 illustrates the RF chain, which consists of
LPF
six RF connectorized components, including a VCO,
a pad/attenuator, a power amplifier (PA), a 3-dB cou-
Figure 4. A diagram of the homodyne radar system. Tx: pler, a low-noise amplifier (LNA), and a mixer. The
transmitter; Rx: receiver. basic system architecture depicted in Figure 4 is very
similar to systems found in many of today’s automo-
signal, and then the distance and velocity of the target tive radars, except those systems are at a much higher
can be estimated: frequency. The total bill of materials (BOM) cost for
the OCW system was US$360, about US$240 of which
cTfr was for the Mini-Circuits RF components. The coffee-
R= (3)
2B can antenna part of the BOM mainly consisted of the
cfd cabling and connectors and was listed as US$54. That
V= , (4)
2fc included the cost for both coffee cans, one of which was
for (fittingly) “Maxwell” House brand coffee. Reduc-
where R is range, V is velocity, and B is the bandwidth ing the cost of all these RF components by moving to
of the modulation frequency. A more extensive discus- a PCB technology with SMT components can signifi-
sion of the signal processing for range and velocity cantly reduce the radar cost because Wi-Fi parts are
measurements for an FMCW radar can be found in [4]. ubiquitous to everyday electronics, from cell phones to
In the OCW system, a breadboard circuit modulator gaming consoles and even home thermostats.
(shown in Figure 1) produces a linear voltage ramp that
causes the voltage-controlled oscillator (VCO) to output “Re-Caffeinated” Coffee-Can Radar System
an FMCW signal. This FMCW signal is transmitted and The first step in redesigning the MIT OCW radar
also used for down converting the received echo. The was to create a system-level analysis of the radar
modulator creates a triangle wave, with a 40-ms period topology so that radar system metrics could be used
and 50% duty cycle, and translates the magnitude of the to determine the performance of the redesign in
ramp to the desired transmit bandwidth. compa­rison to the original system. To start, the OCW
Figure 3 displays a system simulation of a triangle radar system diagram in Figure 4 was implemented
wave and the OCW’s modeled using a Mini-Circuits using behavioral models for the individual RF com-
VCO, showing the voltage and frequency relations with ponents in a communications system simulation
actual part performance and tuning curve pulled from EDA ­software. There are several commercially avail-
its data sheet. It is easy to observe that the VCO fre- able system-design software packages, each offering
quency is not completely linear during the ramp due analysis and RF behavioral block models to support
to this particular VCO’s tuning characteristics. Because the ­system designer. For this work, the RF design and
the same transmit signal with any slight nonlinearity system s­ imulations were implemented in the Visual

64 October 2016
 Original Coffee Can Radar

Mini-Circuits
ZX95-2536C+
Cantenna

5.9 dBm Mini-Circuits Mini-Circuits Mini-Circuits

VAT-3+ ZX60-272LN+ ZX10-2-42+

Mini-Circuits
ZX60-272LN+
Baseband
Signal
Cantenna

Mini-Circuits
ZX05-43MH+

Figure 5. The OCW system diagram implemented with behavioral models.

­ ystem Simulator (VSS) system-design software from

S exact original system topology but replaces the con-
National Instruments (NI). nectorized components’ behavioral model param-
System-simulation blocks are essentially math- eters with performance metrics of comparable SMT
ematical models representing the performance of RF parts, mostly in the US$2.00–3.00 price range. The
components and are configured using information exact part numbers and performance values are
supplied in the vendor data sheets such as gain, 1-dB shown in Figure 6.
compression point (P1dB), and noise figure. The sim- RF system designers have many choices for ampli-
ulator used for this project, VSS, has a built-in radar fiers, mixers, and VCOs in the ISM Wi-Fi bands, avail-
component library that contains more complex mod- able through distributors with convenient online
els such as targets and baseband processing blocks, ordering and quick shipping. For example, Table 1
which were pulled directly onto the system diagram, shows a small sample compilation of parts that were
as shown in Figure 5, representing the OCW transmit/ considered for the PA slot in this redesign project.
receive chain. The system block values for the behav- The decision for each of the active RF functions in the
ioral models were populated in the system simulation redesigned radar was researched in this way before
diagram according to each individual Mini-Circuit the final parts were chosen based on price, size, PCB
component’s data sheet. implementation complexity, and, ultimately, perfor-
Figure 5 shows that, in addition to the transmit/re- mance. Being able to compare the parts’ performance
ceive chain, a target model including distance, velocity, in the system simulator was very valuable to this rede-
and radar cross-section parameters is inserted between sign effort because different parts could be evaluated
the transmit and receive antennas. The system model for system metrics such as total system noise figure,
in Figure 5 supports both RF link budget analysis and transmit power, and margin-to-power compression at
time-domain envelope simulations, which enable base- different points in the system diagram.
band processing algorithms to be applied to the simu- The PCB implementations for the pad attenuator
lated signals. Therefore, the hardware and baseband and low-pass filter (LPF) were chosen to be simple SMT
processing software can be developed jointly by RF capacitors and resistors using the circuit simulator to syn-
and system engineering, providing immediate and di- thesize and design their exact values. Both the LPF and
rect feedback to both engineering groups when design attenuator were designed as π-network circuit topolo-
changes are made. gies so that they could be easily shorted out with a series
The original RF chain in Figure 5 provided the zero-Ω resistor if deemed unnecessary later. Based on
performance baseline for the entire system. The “re- the system simulations, the attenuation was chosen to be
caffeinated” system shown in Figure 6 replicates the 2 dB, and the LPF bandwidth was set to 10 MHz.

October 2016 65
 Redesigned SMT Radar

Maximum
MAX2750 Board Coupler Vivaldi

Board Zero Ohm R

–3 dBm
MAACOM
MAAPSS0075

TriQuint Skyworks
CMY210 SKY67107-306LF

Baseband
Signal Board LPF Vivaldi

Figure 6. The “re-caffeinated” radar design using behavioral models.

Both the attenuator and LPF designs included all the The redesigned coupler function was essentially free
parasitic effects incurred in their SMT components and because it is printed on already-available board space,
mounting layout, which include all distributed effects which saves the cost of the discrete RF component
and full SMT resistor and capacitor models supplied by found in the original system. The performance of the
the component manufactures. Therefore, the frequency coupler for the through insertion loss (S21), coupling
dependency of circuit implementations was fully taken value (S31), and through and coupled port reflection
into account in the system cosimulations both in the RF coefficient (S11 and S33) are shown over a wide band-
link analysis and in the time-domain velocity and rang- width, with their values at the radar’s center frequency
ing extractions. marked on the graph in Figure 7. The 0.19 dB of through
Because the new design was fabricated as a PCB, a loss is quite a savings compared to the 3-dB loss in the
20-dB microstrip coupler was designed using built-in original coupler. The system simulations showed that
distributed coupler circuit models available in the cir- the 20-dB coupled power level was adequate to drive the
cuit simulator. The coupler design was then verified local oscillator (LO) of the mixer chosen for the rede-
using a planar electromagnetic (EM) simulation [Micro- signed system.
wave Office with the AXIEM planar three-dimensional The coupler’s EM simulation mesh is displayed in
(3-D) EM simulator]. These simulators are integrated Figure 8, which shows the meshing of the rounded cor-
within the same NI AWR design environment platform ners and the edge meshing of the traces to accurately
as the VSS, thus allowing the circuit/EM simulation represent the microstrip mode current flow. The EM
results to be used directly within the system analysis. simulation was performed with and without solder

Table 1. A sample parts list for the PA function.

Supporting
Vendor Part Number Gain (dB) P1dB (dBm) Components Price (US$) Power Supply

MAACOM MAAPSS0075 30.0 26.0 16 $1.11 3.3 V @ 275 mA

Avago MGA-412P8 25.5 25.3 13 $1.32 3.3 V @ 200 mA

Microchip SST12LP17E 28 24.0 1 $0.65 3.3 V @ 155 mA

Skyworks SKY65006-48LF 27 24.0 14 $2.33 3.3 V @ 265 mA

66 October 2016
mask layers on top of the FR-4
board, which showed little PCB Microstrip Coupler Design
difference in the total coupler 0 0
performance. The final design Through Loss (L)
used a solder mask window –5 2.45 GHz Coupling Factor (L) –10

Reflection Coefficient (dB)

–0.1892 dB

Through or Coupling (dB)

over the coupler only to em- 2.45 GHz Through Return Loss (R)
phasize the PCB layout (in –10 –26.76 dB –20
other words, it looked cool on Coupled Port Return Loss (R)
the physical board). –15 –30
To further reduce the ove­­­ 2.45 GHz
r­­
all size and improve the –31.09 dB
–20 –40
practicality of the redesigned
radar, the coffee-can antennas
–25 –50
were replaced with the popu- 2.45 GHz
lar Vivaldi planar antennas. –19.98 dB
Vivaldi antennas offer broad- –30 –60
band characteristics, high gain, 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
easy PCB manufacturability, Frequency (GHz)
and good impedance match-
ing to the microstrip feed line. Figure 7. The coupler performance with EM simulation.
A Vivaldi antenna is a surface-
type traveling-wave antenna,
whereby the EM waves travel down the curved path
of the flare along the antenna on the backside substrate
ground. In the region where the separation between
the conductors is small compared to the free-space 2
wavelength, the waves are tightly bound.
As the separation increases over the copper curved 4
backside pattern, the bond becomes progressively 1
weaker, and the waves are radiated away from the
antenna. This occurs when the edge separation is 3
greater than about one-half wavelength. A Vivaldi
antenna is characterized by radiation in the end-fire
direction, with the curvature of the tapered profile Figure 8. The FR-4 20-dB microstrip coupler EM mesh.
having a significant impact on the antenna’s gain,
beamwidth, and bandwidth [5]. Most of the design of
the Vivaldi can be done with a method-of-moments
(MoM) planar simulator, which is very mesh efficient,
runs fast, and is a good fit for this type of planar struc-
ture. AXIEM from NI/Sonnet Software or Momentum
from Keysight are good examples of commercially
available MoM simulators that can be used for this
particular design and antenna optimization. Figure 9
1
shows the Vivaldi antenna’s AXIEM mesh with the
microstrip feed meshed in blue on the topside of the
PCB and the tapered ground plane slot displayed in
purple on the backside.
This Vivaldi design had 1.5–4-GHz bandwidth for
the reflection coefficient, as shown in Figure 10, which Figure 9. The AXIEM mesh of the Vivaldi antenna.
was much more bandwidth than the system needed. The
measured data plotted in Figure 10 shows fairly good
agreement with the EM simulations. Differences in the and save board area and cost by optimizing it for just the
high-frequency measured performance to the modeled 2.45-GHz frequency range.
performance are due to the SMA connector’s influence, To validate the full antenna performance, includ-
which was included in the measured antenna but was ing the effect of the antenna board’s discrete substrate,
not taken into account in the planar EM simulation. Note the designer can use a full 3-D EM finite-element
that this Vivaldi antenna could be made much smaller method (FEM) volume mesh simulator such as ­Analyst,

October 2016 67
 The 2-D pattern cuts in Figure 11 show
Vivaldi Antenna Performance that the Vivaldi has 1.2-dB less directivity,
0 resulting in a broader pattern caused by the
DB(|s(1,1)|)
Vivaldi_Antenna_AXIEM planar nature of the antenna. But the slightly
Reflection Coefficient (dB) –5 DB(|s(1,1)|) lower gain of the Vivaldi antenna is more
Vivaldi_Antenna_IRL_Meas
than compensated for by the additional PA
–10 gain and the lower through-loss in the new
20-dB coupler as opposed to the 3 dB in the
–15 original OCW radar. Additionally, the Viv-
aldi has a significant advantage in terms of
–20 planar form factor in size, cost, ease of assem-
bly, and integration into a system.
–25
1.3 1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9 3.1 3.3 3.5 3.7 3.9 4.1 4.3 4.5 System Simulation RF
Frequency (GHz)
Link Budget Results
After the subsystem components were de-
Figure 10. The performance for the Vivaldi planar antennas, which signed and selected, a system simulation was
replaced the original coffee-can antennas. performed to provide a budget analysis for
the FMCW radar’s power and gain at its cen-
ANSYS’s HFSS, Keysight’s EMPro (all based on FEM), ter frequency of 2.45 GHz. This analysis allows the sys-
or CST’s Microwave Studio (based on finite difference tem integrator to see the contribution of the different
time-domain methods). Figure 11 displays the 3-D radi- individual component’s RF metrics as the signal propa-
ation pattern and a two-dimensional (2-D) cut along gates through the system—a capability extremely use-
boresight for the FEM simulation, including the dis- ful for system engineers who must consider which
crete PCB substrate effects. components need upgrading for higher performance.
Although alternative 3-D EM tools, as previously For this example, Figure 12 shows the comparison
­mentioned, could be used to derive the radiation pat- between the original system (in blue) and the recaf-
terns, this demonstration used Analyst, which sup- feinated system (in green). The OCW system starts
ports the ability for the pattern to be directly read into out with a higher output power from the original VCO
the VSS diagram for system-level analysis. Having the than the new SMT VCO, but as the signal travels to
entire pattern in the system tool allows the designer the transmit antenna, the redesigned system achieves
to vary the θ and φ locations of the target and see the a higher equivalent isotropic radiated power (EIRP).
effect on the system performance metrics when the tar- This is mainly due to the lower insertion loss of the
get is off boresight. For comparison, the original coffee- 20-db coupler and the higher PA gain of the SMT
can antenna was also analyzed in Analyst to provide a replacement part driving the antenna. EIRP is defined
simulated comparison pattern to the Vivaldi. as  the amount of power that a theoretical isotropic

Antenna Pattern Comparison

Mag Max
10 dB Comparison of
Simulated Coffee
Can and Vivaldi
Patterns

Mag 7.3 dB
Ang 90°

Coffee Can Pattern Mag 8.5 dB

Ang 90°
Vivaldi Pattern
10 dB Mag Min
Per Div –30 dB

Figure 11. The Vivaldi planar antenna pattern versus the “cantenna” patterns. Mag: magnitude; Ang: angle; Div: division.

68 October 2016
antenna (one that evenly dis-
tributes power in all direc- 40
tions) would emit to produce 30 Antenna.S4@1
the peak power density 20 30.9 dBm

System Node Power (dBm)

observed in the direction of 10
maximum antenna gain: 0
Antenna.S4@1
–10 20.2 dBm
  EIRP log = PT - L C + G a .  (5)
–20
Radar_Original Coffee Can Radar (dBm)
–30
EIRP and PT (output power of Radar_Redesigned SMT Radar (dBm)
–40
the transmitter) are expressed
in dBm (decibels relative to –50
Target Cross Section = 1m2
milliwatt), cable losses (L c) in –60 Target Velocity = 20 km/h
decibels, and antenna gain –70 Target Distance = 50m
(G a) in dBi [decibels relative –80 At Am Co Tr Ch Re LN M
te Fil
to a (theoretical) isotropic ref- nu pli In up
l An ansm ann An cei A ixer te
at f i ep e r t e i e t e ve r
erence antenna]. The antenna or utr nn t l L nn
os a
a
s
EIRP in the original system Transmit Path Receive Path
is at about 20 dBm, while Radar Component Chain
the new system is around
30 dBm (1 W), which is the lim- Figure 12. A graph comparison of the system budget for the original (blue) and
it for a high-directivity anten- the redesigned (green) coffee-can radar. (The individual data points align with the
na based on U.S. Federal Com- corresponding component description shown below the x-axis.)
munications Commission Part
15 rules for unlicensed wire-
less equipment operating in the ISM bands [6]. runs, the mixer is getting enough LO power to effec-
The path loss, phase change, and Doppler shift of tively perform mixing or to ensure that the LNA is not
the output signal reflected from the target are applied in danger of being compressed.
by the simulator using a target model from the VSS
radar library. The received signal propagates back Estimating Distance and Velocity
through the receive antenna to the LNA and mixer, Once the system simulator performed the budget
where it is down converted to IF. As Figure 12 shows, analysis for RF power and gain, time-domain simula-
the OCW system was receiving about –45-dBm IF tions were run on the same block diagram using the
power RF chain output with the target values shown FMCW triangle-waveform-modulated signal shown
in the graph. in F ­ igure  3, and the frequency (spectral) responses
The redesigned system gets very close to –31 dBm, were plotted using Fourier transformation in the sys-
mostly due to higher gain and power from the selected tem simulator. For this system, the two signals at the
PA (which has higher dc power dissipation as well) in receiving antenna take the form
the transmit chain and an additional 5-dB gain from
the SMT LNA in the receive chain. These system bud- VT sin ^{ T ^ t hh (6)
get calculations indicate that the new system will have
greater range and sensitivity than the original system. from the transmitter and
Similar performance charts can be made for the cumu-
lative noise figure, gain, and compression headroom, VR sin ^{ R ^ t hh (7)
which are all useful system metrics to monitor as the
designer “virtually” designs or selects different com- from the echo delayed by the round-trip echo time.
ponents in the radar’s RF chain. Mixing these signals in the receiver yields
One important value of system-design software
that is often underestimated is its ability to integrate
VT VR sin ^{ T ^ t hh sin ^{ R ^ t hh =
manufacturing tolerances and part performance varia-
tion into the system simulations. The system can then
1 V V # =cos ^{ T ^ t h - { R ^ t hh G .  (8)
2 T R - cos ^{ T ^ t h + { R ^ t hh
be simulated over random performance parameter
variations using the Monte Carlo or corner analysis
techniques so that system engineers can look at system After the high-frequency sum term is removed with
performance variability over assembly runs. This pro- an LPF, the range information is contained in the low-
vides the ability to observe a metric such as power frequency component and is obtained via spectral
headroom to ensure that over long-term ­production analysis of the beat note:

October 2016 69
 { T ^ t h - { R ^ t h . (9)
If the target is stationary, there is no Doppler shift,
and the simulated beat (difference) frequency is due
Figure 13 shows the IF signal spectra after the LPF only to the range of the target. If the target is not station-
in the receiver and illustrates the positive and nega- ary, a Doppler frequency shift will be superimposed
tive frequency peaks resulting from the propagation on the frequency-modulated beat range, resulting in
delay (target distance) and Doppler shift (velocity) in both peaks shifting in the same direction. Analysis of
the FMCW radar signal. The two traces highlight the the beat tone offset from zero is then used to calculate
15-dB improvement in system gain between the old the relative velocity and distance of the target [7]. The
and new systems. target distance and velocity estimates over time were
Once the IF was sampled at the system output, a graphed, as shown in Figure 14.
Fourier transformation was used, and signal process- Both systems shown in Figure 14 were pointed at the
ing was then performed in the frequency domain with same simulated 1-m2 target moving at a relative velocity
both built-in radar system blocks and MATLAB cosim- of 20 k/h at a distance of 50 m. Both systems determined
ulation. The difference between the two peaks can be the exact same target distance with very little noise (see
used to estimate the delay between the reflected signal the lower part of the graph in ­Figure  14). The target’s
and the coupled signal from the transmitter. exact velocity from the Doppler shift was the more dif-
ficult quantity to ­discriminate
from the simulated beat fre-
FMCW IF Power Spectrum quency. This is because most of
0 the shift was due to the change
–20 –0.002734 GHz
in the modulated carrier fre-
–44.91 dBm quency during the propaga-
–40
tion delay (range) compared to
–60 –0.002734 GHz the Doppler related frequency
Target Cross Section = 1 m2
Power (dBm)

–59.96 dBm shift (velocity).

–80 Target Velocity = 20 km/h
However, there was an im­­
Target Distance = 50 m
–100 provement in the velocity accu-
racy determination for the new
–120
system due to the higher re­-
–140 ceived power. Given enough
FMCW Original (dBm)
–160 sampling time, both systems
FMCW Redesign (dBm) average to the same correct
–180 velocity. Still, as the system com­-
–0.1 –0.08 –0.06 –0.04 –0.02 0 0.02 0.04 0.06 0.08 0.1
parison in Table 2 quantita-
Frequency (GHz)
tively shows, the original system
Figure 13. A comparison of the simulated received spectrum for the original and requires a much longer time to
redesigned radar systems. obtain an accurate average veloc-
ity estimation because it varies
much more over time than the
redesign does. This variation
FMCW Radar Simulated Estimations only gets worse with more dis-
50
tant or smaller targets because
Velocity (km/h)

40
the return echo level will be
30 lower. Furthermore, tracking
20 faster moving targets would also
10 yield higher estimation errors in
0 the original design.
FMCW Original Velocity FMCW Original Distance
50.1
Target Distance (m)

FMCW Redesign Velocity FMCW Redesign Distance

The Redesigned PCB
After the system diagram was
50
Target Cross Section = 1 m2 complete, the entire PCB design,
Target Velocity = 20 km/h including RF signal traces and
Target Distance = 50 m SMT components, was laid
49.9
800 1,300 1,800 2,300 2,800 3,300 3,800 4,300 4,800 5,300 5,800 out with a t­wo-layer 31-mil-thick
Time (µs) FR-4 board technology in the
same EDA ­en­­­­­vironment. Using
Figure 14. The time-domain measurements of target distance and relative velocity. a single environment with

70 October 2016
 Table 2. The predicted system comparisons for a target cross section of 1 m2 with 20 km/hr and 50-m distance.

Transmit
Power Cumulative Distance Velocity Total Power
System EIRP IF Power Noise Figure Variation Variation Consumption

Original OCW 20.2 dBm -46.7 dBm 71.0 dB ± 0.02 m ± 21.6 km/hr 0.83 W

Redesigned PCB 30.9 dBm -31.6 dBm 69.7 dB ± 0.02 m ± 8.1 km/hr 1.04 W

the tightly integrated system, circuit, and EM simu-

lators created a design flow for the entire system design
that was much faster and less prone to errors because
changes in one piece of the design (such as the cou-
pler layout) were immediately reflected in its electri-
cal performance and were automatically updated in
the cosimulation system metrics, such as velocity esti-
mations. Therefore, all parasitic effects were incor-
porated into the circuit simulation to create a true
virtual physical design for final verification. Note
that similar capabilities can be achieved with alter-
native software packages. The board was submitted
to a manufacturer along with the BOM for assembly,
resulting in the assembly of the RF core PCB shown
Figure 15. The re-caffeinated OCW radar PCB design
in Figure 15.
(2.5 in × 2.5 in).
The VCO, PA, and LNA layouts were all produced
directly from what was suggested by the manufac-
turer’s data sheets [8]–[10]. As mentio­ned earlier, the inserted through the large holes in the substrate and
2-dB attenuator and the LPF were designed with dis- then moved apart to provide additional transmit-to-
crete SMT components using the Microwave Office receive antenna isolation. The red edging around
circuit simulator, while the AXIEM planar EM tool the radar core is a single 3-D printed case that pro-
was used to design the coupler shown in Figure 15. vides spacing and protection for the backside dc
The total BOM for this 2.5-in × 2.5-in PCB was bias routing on the two-layer PCB.
around US$40 (in low volumes) for the component
counts going onto the board, as listed here: RF Chain Results
•• four SMT RF components The assembled RF PCB in Figure 15 was tested, and
•• 29 SMT capacitors Figure 17(a) displays the output transmit port of the
•• three SMA connectors radar board as a good solid tone with 21.3-dBm out-
•• seven SMT Inductors put power compared to the simulated 23.8 dBm. The δ
•• 11 SMT resistors between the two was attributed to the test-set cabling
•• one PCB 20-dB coupler. attenuation and slightly higher than expected con-
This was much less expensive than the original nector losses. Figure 17(b) shows the down-converted
OCW design, with BOM components that cost IF port signal when a fixed-signal generator was fed
US$250. Further compaction of the FR-4 board could into the receive port with a 2.45-GHz plus 10-MHz
be achieved with little effect to the RF performance.
It is interesting to note that the most expensive com-
ponents in the Figure 15 PCB assembly were the three
RF connectors.
Figure 16 shows the assembled system consisting
of the radar test bed, the Vivaldi antennas for trans-
mit/receive, and the breadboard and power supply
on the backside of the test board. The entire system
fits in the palm of a hand. The antennas were specif-
ically designed on a separate PCB substrate so they
could be moved away from the RF core as needed by Figure 16. The entire radar RF core of the redesigned
the application. In fact, with cabling added, they can system, consisting of the radar test board, Vivaldi antennas,
be mounted together with a pair of wooden dowels and breadboard power supply mounted on the backside.

October 2016 71
 2.45-GHz Transmit Down Converted
Signal 2.45-GHz + 10-MHz Offset
(Approximating Received
Spectrum- Echo Signal)
Analyzed Image

(a) (b)

Figure 17. The measured (a) transmit and (b) IF tones from the radar test board.

a thicker 62-mil FR4 substrate, which we originally se-

lected because it was a more standard and cheaper PCB
option. We found that the PA part was particularly sen-
sitive to the RF grounding under its SMT package, and
the thicker substrate caused it to oscillate. This was di-
agnosed using RF function breakouts created on the test
board, shown in Figure 18, which was fabricated and as-
sembled along with the radar boards. The breakouts al-
lowed the measuring of each individual function, which
was invaluable to de­­bug­ging where exactly the oscilla-
tion issue resided.
We realized on further analysis that the PA oscilla-
Test Board, Version 2 tion could have been predicted early in the design
cycle with just a simple analysis of the manufacturer’s
supplied S-parameter data, as shown in Figure 19. The
Figure 18. The PCB test board with breakouts of each original PA data sheet used an 8-mil FR-4 in the manu-
RF function. facturer’s demo test boards and data sheets, shown as
the blue traces in the graph in Figure 19. A circuit sim-
ulator schematic shown in the top right corner of Fig-
offset to the transmit signal to approximate a mov- ure 19 was used to mathematically add an additional
ing target in the lab. The receive signal is shown to 54-mil substrate height for the four vias under the PA
be successfully down converted (with the tone on the package to represent the additional thickness and
right and its spectrum-analyzer image frequency on grounding inductance of the redesigned system. This
the left). This simple test validates that the core RF was achieved in the circuit simulator by “lifting” the
PCB was close to the system simulator’s predicted S-parameter file’s ground in Figure 17(a) and insert-
performance, works successfully, and can be used as ing the via modifications into the simulated model in
a radar’s RF core. Figure 17(b), resulting in the red traces for gain (S21)
and reflection coefficient (S11 and S22) simulations.
Engineering Lessons Figure 19 clearly exhibits positive return gain
As always, engineering projects never go perfectly marked at 2.506 GHz, which is a very good indicator of
smoothly, and we learned many lessons during the probable oscillations in the actual circuit implementa-
design of the final RF PCB shown in Fig­­ure 16. There tion. Circuit simulations like this take virtually no time
were, unfortunately, two board iterations required in to set up in the simulator and could have saved an en-
this project to achieve successful results for the OCW tire design spin—which, in this case, was a month of
redesign. The first PCB design had an oscillation in cycle time for fabrication, assembly, and test. To solve
the PA. The root cause was traced back to the use of the issue, in the second iteration of the PCB, a thinner

72 October 2016
s­ ubstrate was selected (31 mil)
in conjunction with a few more Simulations with Lifted Ground
35
grounding substrate vias un-
30
derneath the SMT part, which
25

Gain (dB) or Return Loss (dB)

proved stabilize the part in
the simulator. 20
15 Additional Via Inductance
Another interesting but
10 2.506 GHz
common issue we encountered 1.665 dB
was the effect that power sup- 5
plies can have on total RF radar 0
system performance. Figure 20 –5
shows the transmit spectrum –10
of the RF chain when a hobby- –15
ist power supply was used for –20
the RF core. When compared –25
to Figure 17(a), the sideband 1 1.5 2 2.5 3 3.5 4
skirts of the transmit tone can Frequency (GHz)
be seen coming up with dra-
Gain Manufacturer’s Output Return Loss Gain Lifted Output Return Loss
matic shoulders due to the Data Manufacturer’s Data Ground Lifted Ground
additional noise produced by
the inexpensive power supply.
These sideband noise skirts Figure 19. The schematic and simulation of the PA on a thicker RF-4 substrate.
cause very serious degradation
in the radar sensitivity (which
is why military and aerospace companies spend signifi-
cant design effort in creating spectrally pure and low-
noise power supplies and oscillators for their systems).
The original OCW system used batteries, which avoids
this issue totally and produces purer tones, such as the
one in Figure 17.

Conclusions
An entire FMCW radar system was fully designed in
an EDA tool from system concept to layout so as to
facilitate the creation of a new radar RF core having
better performance, lower cost, and a smaller foot-
print and that is more easily mass-produced than the
original MIT OCW coffee-can radar. An EDA system
simulator provided the ability to easily compare the
system performance between the original coffee-can Figure 20. The effect of a cheap power supply on a
design and the new re-caffeinated system for both transmit signal.
RF link and time-domain simulations, as shown in
the various figures and in Table 2. Built-in radar readers are invited to open it up, play with it, redesign
library elements facilitated the building of the target it, and improve the system design. A video of the initial
model and also the baseband processing. version of this application, as presented a 2015 Interna-
A full radar PCB RF core was created, including EM tional Microwave Symposium’s MicroApps session, is
analysis of the planar antennas, and the entire system available at https://youtu.be/DB5TkXgpaW4.
was built and assembled. A system, circuit, and EM
cosimulation encompassed the entire radar, including Acknowledgements
antennas, board coupler, attenuator, LPF, and ampli- We thank John Carroll for assisting with the mea-
fiers and mixer. This cosimulation was used to verify surements and Debra Gomez for helping prepare the
the FMCW radar performance with various target con- graphics in this article.
figurations in a time-domain simulator with baseband
processing performed using both MATLAB cosimula- References
[1] D. Schneider. (2012, Nov. 1). Coffee-can radar: How to build a syn-
tion and built-in baseband processing blocks.
thetic aperture imaging system with tin cans and AA batteries.
For more detail, this design project will be included in IEEE Spectr., vol. 49, no. 11. pp. 24–25 [Online]. Available: http://
a future release of the NI AWR Design Environment, and spectrum.ieee.org/geek-life/hands-on/coffeecan-radar

October 2016 73
 [2] G. Charvat, J. H. Williams, A. Fenn, S. Kogon, and J. S. Herd. Build [5] J. S. Mandeep and M. R. Nicholas. (2008, July). Design an X-band
a small radar system capable of sensing range, Doppler, and syn- vivaldi antenna. Microwaves RF [Online]. Available: http://mwrf.
thetic aperture radar imaging. MIT OpenCourseWare [Online]. com/markets/design-x-band-vivaldi-antenna
Available: http://ocw.mit.edu/resources/res-ll-003-build-a-small- [6] Afar Communication. (n.d.). FCC rules for unlicensed wireless
radar-­s ystem-capable-of-sensing-range-doppler-and-synthetic- equipment operating in the ISM bands [Online]. Available: http://
aperture-radar-imaging-january-iap-2011/ www.afar.net/tutorials/fcc-rules
[3] A. Wojtkiewicz, J. Misiurewicz, M. Nałecz, K. Jedrzejewski, and [7] B. J. Lipa and D. E. Barrick. (n.d.). FMCW signal processing. ­[Online].
K. Kulpa. (n.d.). Two-dimensional signal processing in FMCW Available: http://codar.com/images/about/1990LipaBarr_FMCW.pdf
radars. [Online]. Available: http://staff.elka.pw.edu.pl/~jmisiure/ [ 8] Maacom Technology Solutions. [Online]. Available: http://www.
esptr_base/lect_fmcw/kk97fm.pdf mouser.com/ds/2/249/MAAPSS0075-318038.pdf
[4] E. Hyun and J-H. Lee. (2009, June). Method to improve range [9] Analog Devices. (n.d.) [Online]. Available: http://www.analog.
and velocity error using de-interleaving and frequency inter- com/en/products/rf-microwave/rf-amplifiers/low-noise-amplifiers/
polation for automotive FMCW radars. Int. J. Signal Processing, HMC667.html#product-overview
Image Processing, and Pattern Recognition, vol. 20, no. 2, pp. 11–22. [10} Triquit Semiconductor. [Online]. Available: http://www.mouser.
[Online]. Available: http://www.sersc.org/journals/IJSIP/vol2_ com/catalogviewer/default.aspx?page=522&highlight=772-
no2/2.pdf CMY210&catalogculture=en-US&catalog=648 

Corrections

D
ue to a production error, the illustrations for Department of Mechanical Engineering, University
Table 1 on page 70 of [1] are switched. The of Alberta, Edmonton, Canada. Pedram Mousavi
illustration on the right should appear under (pmousavi@ualberta.ca) is with the Department
the heading “The Gain Measurement of an Antenna,” of Mechanical Engineering, University of Alberta,
and the illustration on the left should appear under the Edmonton, Canada. Fadhel M. Ghannouchi (fadhel.
heading “The RCS ­Measurement of a Tag.” ghannouchi@ucalgary.ca) is with the Department of
On the opening page of [2], the author affiliations Electrical and Computer Engineering, University of
are listed incorrectly. The correct affiliations are as Calgary, Alberta, Canada.”
­follows: “Ramzi Darraji (rdarraji@ucalgary.ca) is with
the Department of Electrical and Computer Engineer­ References
ing, University of Calgary, Alberta, Canada, and the [1] N. C. Karmaker, “Tag, you’re it!,” IEEE Microwave Mag., vol. 17, no.
7, pp. 64–74, July 2016.
[2] R. Darraji, P. Mousavi, and F. M. Ghannouchi, “Doherty goes digi-
tal,” IEEE Microwave Mag., vol. 17, no. 8, pp. 41–51, Aug. 2016.
Digital Object Identifier 10.1109/MMM.2016.2592398
Date of publication: 8 September 2016 

Are You
Don’t miss an issue of this magazine—
update your contact information now!

Moving?
Update your information by:
E-MAIL: address-change@ieee.org
PHONE: +1 800 678 4333 in the United States
or +1 732 981 0060 outside
the United States
If you require additional assistance
regarding your IEEE mailings,
visit the IEEE Support Center
at supportcenter.ieee.org.
© ISTOCKPHOTO.COM/BRIANAJACKSON

74 October 2016