The present invention relates to phase detectors, and in particular, to a method for generating a differential signal representing the phase difference between two input signals, and to a phase detector operating at high frequencies.
In long distance transmission systems operating at high bit rates over standard signal fiber lines, data receivers may receive significantly distorted signals. Inter-symbolic interference, finite bandwidth, fiber nonlinearity and other non-idealities increase the probability of erroneous recognition of a received bit. For these reasons, it is often necessary to place, along the transmission line, data regenerating channel systems that sample a received signal and retransmit it to either a successive data regenerating system or to the end receiver.
The incoming data at the receiver may be considered as a varying analog signal from which a synchronization or clock signal may be recovered. Recovering the clock in the form of a signal that generally oscillates between a higher level and a lower level signal from the incoming signal is essential for sampling it correctly to regenerate the digital data to be transmitted.
Of course, the clock signal could alternatively be transmitted together with the data stream, and the clock can be easily filtered at the receiver. In the majority of cases, the clock must be recovered from the data stream using a phase locked loop (PLL).
The phase detector PD is input with the digital signal DAT to be regenerated and retransmitted, and the recovered clock CK. The phase detector PD commonly includes a differential stage that outputs a differential signal OUT+, OUT− representing the phase difference between the digital signal DAT and the clock CK. This differential signal is produced by comparing the transition edges of the digital signal and the clock signal.
The loop filter LP is input with the differential signal OUT+, OUT− and generates a control voltage Vc for a voltage controlled oscillator VCO by low pass filtering the differential component of the differential signal OUT+, OUT−. If the control voltage Vc is not null, the VCO adjusts the frequency of the recovered clock CK until the control voltage becomes null.
If the digital signal DAT switches regularly, the phase detector is able to continuously compare the transition edges of the recovered clock CK and the signal DAT. Thus, the recovered clock has a good precision. Differently, when the digital signal is a non-return to zero (NRZ) signal, such as the one depicted in
Many types of phase detectors are available. It is worth mentioning that the classical phase and frequency detector (PFD), the bang-bang detector and the linear phase detector are frequently used.
The PFD detector, shown in
In this mode the flip-flops generate two output pulses. The difference between the duration of these two pulses represents the phase error between the two input signals. The advantage of this detector is its capability of sensing both phase errors and frequency errors, and that its output is proportional to the phase mismatch. A second advantage is that when the two inputs are synchronized, the duration of the output pulses is null and there is no injection into the loop filter, and as a consequence, the litter is minimized. A disadvantage of this architecture is that it does not work when there is an absence of transitions in the input signal, and so it is not usable for regenerating data for a NRZ transmission system.
A possible approach to overcome this limitation is represented by the so-called bang-bang phase detector, the working principle of which is illustrated by the timing diagram of
The disadvantage of this phase detector is that its output is not proportional to the phase error between data and the clock, i.e., this phase detector has a non-linear transfer function. A system for regenerating data that employs a bang-bang phase detector may continuously oscillate between a phase lead and a phase lag. This increases the frequency jitter of the recovered clock.
Another family of phase detectors is represented by the linear phase detectors like the Hogge phase detectors, which generate a signal proportional to the phase difference of their input signals. Both linear and bang-bang phase detectors exploit a similar working principle, which is as follows. At the transition of the incoming data, a positive or negative current or voltage pulse is output toward the loop filter, depending on whether the data leads or lags the clock. The amplitude of the pulse may be constant (bang-bang phase detectors) or proportional (linear phase detectors) to the phase difference between the data and the clock, as disclosed in the article by Aaron et al., titled “Integrated Fiber-Optic Receivers”, Kluwer Academic Publishers. Unfortunately, it is very difficult to use them when the data rate is relatively high because they are based on the use of flip-flops, which require a certain time for generating a stable output.
An object of the present invention is to provide a fast and relatively straightforward phase detector which may be used in a phase-locked loop, and is particularly suited for operating at relatively high data rates based on the use of fast linear (analog) stages.
This and other objects, advantages and features in accordance with the present invention are provided by a phase detector receiving as input a generally oscillating signal, for example a distorted digital data signal, and a clock signal for outputting a differential signal representing the phase difference between the oscillating signal and the clock signal.
The phase detector comprises a first differential pair of transistors respectively driven by the clock signal and by its inverted replica for generating a differential signal corresponding to the currents respectively flowing in the transistors of the first differential pair. At least one auxiliary differential pair of transistors is respectively driven by the generally oscillating signal, and its inverted replica having its common current node is coupled to corresponding current nodes of the first differential pair. A current generator may bias all of the differential pairs.
The output differential signal is non-null only when there is a transition and there is a phase difference. For the time the oscillating input signal does not undergo any transition, the differential signal may remain null.
Although the novel phase detection of this invention is outstandingly fast, as in known phase detectors, if there are long periods of time during which the generally oscillating signal does not switch, the precision of the frequency of the recovered clock may progressively worsen.
According to a preferred embodiment of the invention, the above mentioned problem is addressed by providing the phase detector with a feedback loop that regulates the current generated by the biasing current generator. The loop includes a sensor that monitors the transition density of the generally oscillating input signal, and increases the bias current of the differential transistor pairs when the transition density decreases. The amplitude of the output differential signal increases because of the increased gain of the differential stage, thus making the VCO that is present downstream adjust more promptly the frequency of the recovered clock.
It has been found that the transition density of the oscillating input signal directly affects the time average of the common mode current of the output differential signal, and that an effective feedback loop may be formed by using a sensing circuit of the output common mode current for generating a voltage representative of the transition density of the input oscillating signal, and a correction circuit including an amplifier for amplifying a difference between the representative voltage and a reference value. The feedback loop regulates the gain of the differential pairs to make null this difference.
The different aspects and advantages of the invention will appear even more evident through a detailed description of few embodiments and by referring to the attached drawings, wherein:
A first embodiment of a phase detector of the invention is depicted in
In the ensuing description reference will be made to a digital signal DAT, but the same considerations hold for any other generally oscillating signal, such as a sine waveform, a saw-tooth signal and the like. The amplitudes of the signal DAT and the recovered clock CK may be chosen such that when the digital signal DAT is not switching, the second differential pair Q1, Q2 draws the whole current of the generator, while the first differential pair Q3, Q4 does not deliver any output current to the loop filter. For a digital signal DAT switching between a positive +V and a negative voltage −V, this condition is satisfied if the absolute value V is always greater than the maximum absolute voltage level of the recovered clock.
The above mentioned condition is not strictly necessary. As may be easily noticed by an expert technician, the clock amplitude may even surpass the absolute value V of the oscillating input signal if the transistors Q3, Q4 of the first differential pair are provided with appropriate emitter degeneration resistors (not shown in
During data transition, because of finite rise and fall times of the digital signal DAT, if the recovered clock is not synchronous with the signal DAT, there will be a time interval during which the differential pair Q3, Q4 absorbs a portion or the whole constant bias current Ipd, as illustrated in
If at a transition of DAT (DATN) the clock signal CK is higher than CKN, then the current OUT+ is greater than the current OUT−. The opposite situation occurs when at the transition of DAT (DATN) the clock CK is lower than CKN.
By assuming that, because of finite rise and fall times of the oscillating input signal DAT and of its inverted replica DATN, there is a time interval T1 in which a part of the bias current Ipd is absorbed through the transistors Q3 and/or Q4 and that half of the bias current Ipd is absorbed through the differential pair Q1, Q2 and half through the output differential pair Q3, Q4. The total electrical charge flowing in the differential pair Q3, Q4 will be
Ipd*T1/2.
This charge transfer splits itself between the two transistors Q3 and Q4 according to the well known hyperbolic tangent function that characterizes every differential pair. If the phase mismatch between the recovered clock CK and the input signal DAT is relatively small, it is possible to approximate this hyperbolic function with a linear function. Thus, the differential output signal will be approximately proportional to the phase mismatch.
If the bases of the transistors Q3 and Q4 are at the same voltage during a transition of the signal DAT, that is, if the clock is perfectly synchronous with the digital signal DAT, both transistors absorb only a common mode current. This is while the differential mode current, which represents the output of the phase detector, is null.
Should the base voltage of the transistor Q3 be higher (lower) than the base voltage of the transistor Q4 during a transition of the signal DAT, that is, if the recovered clock CK leads (lags) the signal DAT, then a greater (smaller) current will flow in Q3 than in Q4. In this case the phase detector will output a non-null differential signal because the two input signals are out of phase.
The fact that the amplitudes of the output current pulses OUT+ and OUT− of the phase detector of the invention are practically proportional to the delay between the clock CK and the signal DAT, this makes the phase detector particularly suited for realizing phase-locked loops that are capable of recovering accurately the clock from a NRZ data stream. In fact, the phase detector of the invention outputs a null differential signal in the case of phase matching between the input signal DAT and the recovered clock, thus minimizing the frequency jitter of the recovered clock CK.
The phase detector of the present invention may work at very high bit rates (>10 Gb/s) because it is substantially composed of four transistors, which may be either bipolar junction transistors (BJT) or MOSFETs, with relatively short recovery times. These transistors may switch at extremely high frequency.
Moreover, the phase detector of the invention is ideally suited also for NRZ digital input signals, because it does not generate spurious outputs corresponding to missing transitions. In fact, as long as the signal DAT does not switch, the differential pair Q3, Q4 that generates the differential signal OUT+, OUT− remains unable to draw any current from the bias current generator Ipd. This is because the second differential pair Q1, Q2 absorbs the whole bias current of the common current generator.
According to an alternative embodiment, there may be two auxiliary differential pairs Q1, Q2 and Q1′, Q2′ coupled, respectively, to the collector nodes of the transistors Q3, Q4 of the first differential pair according to the circuit diagram of
In addition, output transistors Q5, Q6 are respectively connected in series to the collector nodes of the first differential pair and are both controlled by a control voltage REF for keeping the output transistors Q5, Q6 in a conduction state, at least and preferably only during the transitions of the input signal DAT. This may be ensured simply by choosing a control voltage REF between the maximum and the minimum values of the oscillating signal DAT.
Let us suppose that the oscillating input signal DAT is a digital signal, such as that of
To reduce the loss of precision that may be caused by a decrease of the transition density, a phase detector with a variable gain is employed and the gain is increased as the transition density decreases. In this way, the VCO downstream of the phase detector receives a control voltage Vc of enhanced amplitude and adjusts more promptly the frequency of the recovered clock.
Accordingly, the phase detectors of the invention depicted in
As noted, several sensing means or circuits for detecting the transition density are known and may be formed, for example, by a counter that counts clock pulses between successive transitions of the oscillating input signal, and by a circuit that generates a signal V2 representative of a time average of these counts.
According to one aspect of a phase detector of the invention, at least a differential stage for outputting a differential current OUT+and OUT− representative of the phase difference between the oscillating input signal DAT and the recovered clock CK is used. This sensing circuit may be implemented in a straightforward manner by generating a representative signal Vs as a function of the time average of the output common mode current.
It has been found that, when the input signal does not switch (oscillates), the differential output signal is null. Thus, the time average of the output common mode decreases as the transition density decreases.
Preferably, the signal V2 representative of the transition density of the oscillating input signal DAT is obtained by low pass filtering the common mode component of the differential output signal. This signal V2 is compared with a reference value V1, and the gain of the phase detector is regulated in a feedback mode to make the signal V2 equal to V1 by regulating the bias current of the differential pair that generates the output differential signal OUT+, OUT−.
Accordingly, a preferred embodiment of the phase detector of the invention is depicted in
The regulation loop is implemented by adding a third differential pair of transistors Q3′, Q4′ that may be identical or scaled replicas of the transistors Q3, Q4 of the first (output) differential pair. These transistors are similarly driven by CK and CKN, such that the currents flowing in the transistors Q3′ and Q4′ are equal or proportional to the currents flowing in the corresponding output transistors Q3, Q4 of the first differential pair. The output common mode current flowing in the differential pair Q3′, Q4′ is forced through a low pass filter R2, C2, for generating a voltage V2 representative of the time average of the output common mode. current of the phase detector, and thus of the transition density of the input signal.
The voltage V2 is applied to a first input of an error amplifier G. The other input of the error amplifier G receives a reference voltage V1 that may be obtained by forcing a reference current Iref through a resistor R1. The error amplifier G regulates the current Ipd generated by the common bias generator for all three differential pairs to make V2 equal V1.
In periods of time during which the input digital signal DAT ceases to switch, the voltage V2 on the low-pass filter R2, C2 decreases. This signals that the time average of the common mode current forced through the filter is diminishing. The high gain differential error amplifier G input with the voltages V1 and V2 regulates the current Ipd that biases all three differential pairs of transistors to make null the difference between V2 and V1.
When a transition occurs after a long sequence of substantially equal input values, the transistors Q3 and Q4 of the first differential pair are biased with a relatively enhanced bias current. The gain of the differential stage is at a correspondingly enhanced level.
Even the phase detector of
| Number | Date | Country | Kind |
|---|---|---|---|
| 03425274.2 | Apr 2003 | EP | regional |