0% found this document useful (0 votes)
40 views

Costas 2

This document presents a time domain model for a Costas loop-based QPSK receiver for millimeter wave wireless communication systems operating at around 100GHz. It describes the operation of the four-phase Costas loop and its components. The model is implemented in Matlab and simulates the dynamic behavior of the Costas loop in the presence of noise, linearity limitations, and bandwidth constraints. Simulation results are presented and compared to the frequency domain transfer function derived from the Costas loop parameters.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
40 views

Costas 2

This document presents a time domain model for a Costas loop-based QPSK receiver for millimeter wave wireless communication systems operating at around 100GHz. It describes the operation of the four-phase Costas loop and its components. The model is implemented in Matlab and simulates the dynamic behavior of the Costas loop in the presence of noise, linearity limitations, and bandwidth constraints. Simulation results are presented and compared to the frequency domain transfer function derived from the Costas loop parameters.
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Time Domain Model for Costas Loop Based QPSK

Receiver
Maarten Tytgat, Michiel Steyaert and Patrick Reynaert
K.U. Leuven ESAT-MICAS
Kasteelpark Arenberg 10
B-3001 Leuven, Belgium
[email protected]

Abstract—A complete Matlab model is made for a millimeter are shown of the complete system with realistic quantities for
wave wireless communication system including a four-phase data communication around 100GHz with Gbit/s data rates.
Costas loop for carrier recovery and QPSK demodulation.
Simulation results are presented to demonstrate the dynamic
behaviour of the Costas loop and the effects of noise, linearity and II. T HE QPSK COSTAS LOOP
bandwidth limitations. The time domain results are compared to
the frequency domain transfer function derived from the Costas
A. Operation
loop parameters. The Costas loop was first proposed by J. Costas as a phase
tracker for (suppressed-carrier) AM signals [3]. It was later
I. I NTRODUCTION
modified to demodulate QPSK and MPSK signals [4], [5].
The continuous demand for increasing data rates has pushed The circuit diagram of the four-phase Costas loop is shown in
the operating frequency of wireless communication systems Fig. 1a. Suppose the RF signal VRF is QPSK modulated:
into the millimeter wave region. Larger bandwidths are thus
available, allowing higher data rates with less complicated VRF = I(t) sin (ωt + θ) + Q(t) cos (ωt + θ) (1)
modulation schemes such as ASK, BPSK and QPSK.
An important issue when dealing with these high frequencies Where I(t) and Q(t) can be ±1, varying at the symbol
and high data rates is the problem of carrier recovery. Coherent rate. Amplitudes are disregarded for simplicity. This signal
detection requires a good phase and frequency reference with is multiplied with an LO with the same frequency and an
respect to the carrier of the RF signal. instantaneous phase of θ′ and a 90◦ phase shifted version of
If carrier recovery has to be done at baseband, high speed this LO. Then after low pass filtering the signals are:
A/D-converters are needed, increasing power consumption and
ZI (t) = I(t) cos φ − Q(t) sin φ (2)
design complexity [1].
The carrier recovery can also be performed in the analog ZQ (t) = I(t) sin φ + Q(t) cos φ (3)
domain. Three important techniques are known: differential
With φ = θ − θ′ . If we assume that the pahse error is always
demodulation using a delay line to mix with a previous
|φ| < 45◦ , the output of the limiters is:
symbol, frequency multiplying and the Costas loop.
The problem with a delay line is that it has to provide a delay LI (t) = I(t) (4)
of one symbol period. For data rates in the order of several
LQ (t) = Q(t) (5)
Gbit/s, that would mean delay lines with lengths in the order
of tens of millimeters. Furthermore, the data rate is fixed by The error signal is then:
the length of the delay element [2].
Using frequency multiplying would require circuits working at ǫ(t) = ZQ LI − ZI LQ
twice the carrier frequency for BPSK and four times the carrier
= I 2 (t) sin φ + Q(t)I(t) cos φ
frequency for QPSK, which is unrealistic for frequencies
exceeding 100GHz. − I(t)Q(t) cos φ + Q2 (t) sin φ
The Costas loop is therefore a suitable candidate for carrier = 2 sin φ
recovery and demodulation at millimeter wave frequencies, as ≈ 2φ (for very small φ) (6)
already demonstrated in [1].
This paper presents a Matlab model and simulation results of This way, a phase error signal is obtained, which can adjust
a wireless communication system operating at 100GHz. Time the VCO in order to maintain phase and frequency lock. If
domain simulations allow to analyze the dynamic properties the absolute phase error is initially larger than 45◦ , the loop
and the behavior in the presence of noise of the Costas loop. will still lock, but the received constellation will have a fixed
The paper starts with the operation of the four-phase Costas rotation of a multiple of 90◦ with respect to the transmitted
loop and the discrete time equivalent. Then, simulation results constellation.
Fig. 2. Overview of the simulated wireless data communication link.

TABLE I
C OSTAS LOOP DESIGN PARAMETERS

Carrier frequency fc 100GHz


Symbol rate R 4GBd/s
(a) Data rate D 8Gbit/s
Phase constant KP 0.75V/rad
VCO gain KV CO 5GHz
Arm filter pole farm 8GHz
Loop filter pole floop 200MHz
RF bandwidth B 10GHz
Distance d 10cm
Free space path loss FSPL 52.4dB
Transmitted power PT x 0dBm
Received power PRx −52.4dBm
Noise Figure LNA NFLNA 10dB
SNR at output LNA SNRLNA 11.5dB

The channel is modelled with frequency and distant dependent


free space path loss. The antenna gains are assumed to be zero.
(b) White noise is added at the receiver antenna as a gaussian
q
Fig. 1. Schematic of the four-phase Costas loop (a) and discrete time Fs
distributed random number with σ = kT 2 (Fs is the
equivalent (b).
sampling frequency). The total noise power is determined by
the input bandwidth B.
B. Discrete time implementation The receiver features an LNA with a gain G, noise figure NF
and linearity specified by P1dB .
In order to make a model of the Costas loop in Matlab, the The design parameters of the Costas loop are the coefficients
continuous time circuit has to be transformed to the discrete of the arm filters and loop filter and the gains of the different
time domain. This is shown in Fig. 1b. In this schematic, the blocks. In the following simulations, the arm and loop filters
variable t[k] is the sampled time and Ts is the sampling period. are implemented as first order low pass filters. The system thus
The VCO is represented by integrating the output of the loop resembles a second-order PLL, with a loop gain of KP KV CO .
filter and multiplying by KV CO to get the instantaneous phase By definition, KP = φǫ is the phase constant of the Costas loop
θ′ for the sin and cos functions. and it is dependent on the gains within the Costas loop and
This discrete-time model can be directly translated into Matlab the amplitude of the received signal.
code. The filters are implemented as IIR filters using difference The used parameters in the reported simulations are shown in
equations. table I. These numbers are realistic values for a transmitter and
receiver in a modern CMOS technology at a carrier frequency
III. S IMULATION OF COMMUNICATION LINK
around 100GHz.
A. System overview
B. Simulation results
In order to reflect a realistic situation, the complete
transmitter-receiver system of Fig. 2 was translated into Matlab To illustrate the phase detection mechanism of the four-
code and simulated in the time domain. The transmitter can be phase Costas loop, an open loop simulation is performed by
configured for a certain carrier frequency fc , symbol rate R setting KV CO = 0. The relevant signals are plotted in Fig. 3.
and output power PT x . Additionally, certain non-idealities can Noise has not been included in this simulation.

be added, such as phase and frequency steps and frequency The carrier is given a phase difference of 22.5 (π/8) with
drift. respect to the VCO. From the bottom right graph, the phase
LI ZI LQ ZQ 100.2

100.15

100.1

f (GHz)
1 1

0 0
100.05 fc
−1 −1 instantaneous fVCO
100

99.95
0 2 4 6 0 2 4 6 0 2 4 6 8 10 12 14 16
t (ns)
ZQ.LI ZI.LQ ε = ZQ.LI−ZI.LQ

0.6 0.5 VCO control Fig. 5. Response to a frequency step of 100MHz.

same way as in a second order PLL [6].


0 0
To compare the time domain simulation results to the fre-
quency domain predictions, simulations are performed with
analog phase modulation on top of the QPSK modulation. The
−0.6
0 2 4 6
−0.5
0 2 4 6 θ from formula 1 is varied at the modulation frequency fm ,
t (ns) t (ns) with an amplitude |θ|. The Costas loop will suppress force
the VCO to have the same phase modulation, thus making |φ|
Fig. 3. Signals in the Costas loop for a phase difference of π8 between carrier zero. This will only succeed for low modulation frequencies.
and VCO. The loop is cut in order to observe the phase dection mechanism. With increasing modulation frequency, the magnitude of the
phase error |φ| is compared to the phase error transfer E(s)
LI ZI LQ ZQ function which is derived from the parameters of the Costas
loop:
φ(s) s
E(s) ≡ = 1 − H(s) = (7)
1 1 θ(s) s + KV CO KP F (s)
0 0
As can be seen in Fig. 6, the time domain results show good
−1 −1
agreement to the frequency domain predictions.
With the inclusion of noise, the received constellation looks
0 2 4 6 0 2 4 6 like Fig. 7. This is a plot of the signals ZI and ZQ , i.e. before
the limiters, sampled at the data rate. This result is obtained for
ZQ.LI ZI.LQ ε = ZQ.LI−ZI.LQ
a received signal strength of −52.4dBm and SNR of 21.5dB.
0.6 0.4 VCO control
Behind the LNA with a noise figure of 10dB, which is a
0.2 realistic number in CMOS at 100GHz, the SNR is 11.5dB.
The corresponding eye diagram of the ZI signal is plotted in
0 0 Fig. 8.
The bit error rate can be simulated in function of the SNR at
−0.2
the output of the LNA. This is shown in Fig. 9. The noise at the
−0.6 −0.4 input is increased, while the signal level is held constant so as
0 2 4 6 0 2 4 6
t (ns) t (ns) not to change the KP of the loop. For larger SNRs than 3.5dB,
the simulation time needed to encounter a sufficient amount bit
errors becomes too long. The typical waterfall curve is clearly
Fig. 4. Signals in the Costas loop for an initial phase difference of π8 , this observed.
time with the loop closed. The VCO frequency is adjusted so that the phase
error is eliminated.
IV. C ONCLUSION
A complete time domain Matlab model has been presented
detector constant can be derived: KP = 0.3Vπ/8 ≈ 0.75V /rad. for a wireless communication system at millimeter wave
The loop filter eliminates the spikes on the ǫ signal. frequencies, using a four-phase Costas loop as carrier recovery
When the loop is closed (KV CO = 5GHz/V), the phase circuit and QPSK demodulator.
tracking can be observed (Fig. 4). The operation of the phase tracking mechanism has been
The response of the loop to a frequency step of 100MHz illustrated and the equivalence to a second-order PLL has been
is shown in Fig. 5. It shows the carrier frequency and the shown with simulations.
instantaneous VCO frequency. The maximum frequency step By using realistic numbers, an idea is given about the quanti-
that can be allowed is determined by the loop dynamics in the ties involved in high data rate wireless communication around
−1
10

10

−2
0 10

BER
−10
E(s)
Magnitude (dB)

−20 |φ|/|θ| −3
10

−30

−40
−4
10
−2 −1 0 1 2 3 4
SNRLNA (dB)
−50

−60 7 Fig. 9. Simulated Bit Error Rates in function of SNR at the LNA output.
8 9 10 11
10 10 10 10 10
Frequency (rad/s)

100GHz.
Fig. 6. Error transfer function E(s) and time domain simulations of phase
error. The influence of noise is shown in the form of a constellation
diagram, an eye diagram and a BER versus SNR graph.
This model can be used to predict and optimize the perfor-
mance of a Costas loop based receiver.
ACKNOWLEDGMENT
This research is partly supported by the ERC Advanced
1
Grant 227680 (DARWIN).
R EFERENCES
[1] S.-J. Huang, Y.-C. Yeh, H. Wang, P.-N. Chen, and J. Lee, “An 87GHz
Q

0
QPSK transceiver with costas-loop carrier recovery in 65nm CMOS,” in
Solid-State Circuits Conference Digest of Technical Papers (ISSCC), 2011
IEEE International, feb. 2011, pp. 168 –170.
[2] H. Takahashi, T. Kosugi, A. Hirata, K. Murata, and N. Kukutsu, “10-
−1 Gbit/s quadrature phase-shift-keying modulator and demodulator for 120-
GHz-band wireless links,” Microwave Theory and Techniques, IEEE
Transactions on, vol. 58, no. 12, pp. 4072 –4078, dec. 2010.
−1 0 1 [3] J. Costas, “Synchronous communications,” Proceedings of the IRE,
I vol. 44, no. 12, pp. 1713 –1718, dec. 1956.
[4] C. Weber and W. Alem, “Demod-remod coherent tracking receiver for
Fig. 7. Constellation plot for SNR = 11.5dB at output of LNA. The data QPSK and SQPSK,” Communications, IEEE Transactions on, vol. 28,
rate is 8Gbit/s. no. 12, pp. 1945 – 1954, dec 1980.
[5] H. Osborne, “A generalized ”polarity-type” costas loop for tracking
MPSK signals,” Communications, IEEE Transactions on, vol. 30, no. 10,
pp. 2289 – 2296, oct 1982.
[6] F. M. Gardner, Phaselock techniques. John Wiley and Sons, July 2005.
Z
I
1.5

0.5
Amplitude

−0.5

−1

−1.5
0 1 2 3 4 5 6 7
Time (s) x 10
−10

Fig. 8. Eye diagram of the ZI signal. The data rate is 8Gbit/s.

You might also like