Designing coherent carrier-recovery circuitry for demodulating suppressed-carrier phase-shift-keying (PSK) signals involves several trade-offs and performance considerations. A number of approaches is available, although this article will focus on a design for a multimission PSK demodulator which can accommodate varying data rates in different PSK modulation schemes without requiring any change in configuration. Such a demodulator is ideal for satellite ground stations receiving data from various remote-sensing satellites of different payload characteristics.

Figure 1 shows a simplified configuration for the PSK demodulator. It consists of an input automatic-gain-control (AGC) amplifier, coherent carrier-recovery circuitry, and coherent detector circuitry. Intermediate-frequency (IF) signals plus noise are bandpass filtered, boosted through the AGC amplifier, and routed in parallel to the carrier-recovery circuit and the coherent data detector. The carrier-recovery circuit regenerates the coherent reference for demodulation that is routed to the data detector. The coherent data detector extracts in-phase (I) and quadrature (Q) data streams, which are lowpass, filtered and fed to the corresponding plug-in of the bit-synchronizer and signal-conditioner (BSSC) unit. The BSSC unit recovers the coherent symbol timing for synchronizing the data to the symbol timing clock. In this case, the BSSC unit supplies serial data and clock outputs.

Most applications can be satisfied by one of the three types of carrier-recovery configuration: a multiplication loop (such as a squaring loop for BPSK), a Costas loop, and a remodulator loop.^1,2 Other types of carrier-recovery schemes are extensions or modifications of these techniques. For example, a multiplication loop for MPSK (Fig. 2) makes use of an Mth-order nonlinear square-law function preceded by a bandpass filter to remove the modulation.

A conventional PLL, operating at a frequency of M × f_c, where M is the harmonic multiplier and f_c is the carrier frequency, locks to the Mth-harmonic component of the nonlinear output while the voltage-controlled oscillator (VCO) is divided by M to provide the desired reference carrier frequency.

In a BPSK Costas loop (Fig. 3), an estimate of carrier phase is obtained by multiplying (with two phase detectors) the input suppressed carrier plus noise with the output of the VCO and a 90-deg.-shifted version of the VCO's signal, respectively, filtering the results of the two multiplications, and using the product of the two filtered signals to control the VCO's phase and frequency. When the filters in the I and Q arms are controlled by integrate-and-dump circuits, the loop is called a Costas loop with active arm filters.^3-6

An optimal phase estimator requires a hyperbolic tangent [tanh (K E_b/N₀)] nonlinearity following the I-arm filter. For large values, tanh(x) equals the polarity or sign of x(±1), and can be implemented with a hard limiter.⁷ An optimum loop has been practically realized by approximating this hyperbolic tangent nonlinearity.⁸ This type of loop is called the hard-limited or polarity loop (Fig. 4). The inclusion of a limiter introduces a signal suppression factor that can improve or degrade the tracking performance.⁶ Also, the limiter allows the replacement of the analog third multiplier, which yields the loop error signal, with a chopper-type device that typically exhibits much less DC offset or DC drift instability.⁹

In a hard-limited modified Costas loop, a quadrature arm filter is removed to improve the acquisition characteristics.¹⁰ The modified Costas loop generates a frequency-restoration force approximately proportional to the frequency error at the output of the third multiplier. During acquisition, the loop acts like an automatic-frequency-control (AFC) circuit with a bandwidth proportional to the square of the Costas loop bandwidth. This modification allows acquisition for frequency offsets greater than the loop bandwidth. Pull-in is achieved from initial frequency errors compatible to the cutoff of the lowpass filter in the I channel. Unfortunately, a trade-off is degradation in the loop tracking jitter.⁶

The hard-limited modified Costas loop achieves significant improvement in false-lock performance, at the cost of increased tracking jitter.¹¹ The false-lock phenomenon arises in Costas loops while acquiring suppressed carrier signals (rather than unmodulated carriers), due to the distortion created by the finite bandwidths of the I and Q filters. In essence, the incoming data pulses in the I and Q channels are no longer ideally rectangular and, when combined, may produce a DC control term when the received signal is offset from the VCO by an integral multiple of one-half the data rate (data-related sidebands).^12,13

The false-lock state corresponds to the condition whereby a stable null point exists in the loop's error characteristic as a function of phase (the so called "S-curve" of the loop) when the receiver's local VCO is offset in frequency from that of the received signal in noise.¹¹ Another modified Costas loop configuration, where the sum of the squares of the I and Q channels is processed to contribute to the loop error signal, may inhibit stable operating points at false lock.¹³

A conventional quadrature Costas loop has been used for the carrier extraction of QPSK signals.⁹ However, the most popular approach uses a modified Costas loop with loop crossover arms (Fig. 5).^14,15 The sign of the output of the arm filters produced by the limiter is used to crossover and mix with the signal from the opposite arm. The limiters effectively demodulate the QPSK quadrature bits and the crossover produces a common phase-error term that is cancelled after subtraction. The subtraction leaves a remainder error term that is used to generate an error signal for phase control of the loop VCO, thereby closing the loop.

Remodulation is another popular carrier-recovery technique (Fig. 6). The incoming signal is demodulated and the message waveform is recovered. This baseband waveform is used to remodulate the incoming signal; if the waveforms are rectangular and time aligned, the remodulation procedure removes the modulation completely. The output of the balanced modulator has a pure carrier component at the input frequency and the PLL tracks the component. The remodulator is stochastically equivalent to the polarity loop, i.e., the hard-limited Costas loop.¹⁶ The remodulator, however, is typically implemented at low frequencies (below IF) and cannot be used for multiple data rates due to time delays which impact the realization of a wide-band synchronizer.¹⁷