WO2007015191A1

WO2007015191A1 - Delay-locked loop

Info

Publication number: WO2007015191A1
Application number: PCT/IB2006/052550
Authority: WO
Inventors: Pavel Petkov; Jim Conder; Friedel Gerfers
Original assignee: Nxp B.V.
Priority date: 2005-08-03
Filing date: 2006-07-25
Publication date: 2007-02-08
Also published as: ATE532267T1; CN101233689A; EP1913696A1; JP2009504058A; EP1913696B1

Abstract

The delay-locked-loop (100) comprises at least two delay elements, of which a first delay element (10) has a positive delay line and an input for receiving a clock, and of which a second delay element (11) has a negative delay line and an input for receiving a clock, a clock selector (13) for selecting the clock from one of the two delay lines, a phase detector (14) with an input for receiving data and for comparing the phase of the data to that of the selected clock, and a control block (12) which produces control signals for controlling the two delay lines such that they react in opposite directions to a signal from the phase detector (14). Such a delay-locked-loop suitable for accurate clock generation in plesio-sinchronous communication systems.

Description

DELAY-LOCKED-LOOP

The invention relates to a delay locked loop.

Delay locked loops are known from the general state of the art.

In modern high speed communications it is often necessary to align a clock to an incoming signal. This may be done by delaying the clock using a delay- locked-loop (DLL) locked to transitions in the data stream.

Such delay-locked-loops have a limited operational range (minimum and maximum delay and operating frequency) which limits the amount of clock-to-data skew that can be compensated. This in turn places tight requirements upon the data being received (often not possible in high data rate applications). To overcome this, several methods for "infinite range" delay have been proposed. These generally involve using two or more separate delay-locked-loops, each with its own control, start-up, range and stability requirements.

The simple analogue delay-locked-loop which is known from the prior art has a minimum delay determined by its physical (process technology) parameters and a maximum delay determined by the frequency of the signal being passed.

For long delays at high frequency, very long delay chains are required. This means high sensitivity of the delay line (delay per unit control signal) and this introduces jitter, which in turn restricts the maximum useful delay. A typical delay- locked-loop has a useful compensation range limited to about 2 clock periods. The initialisation of a prior art delay-locked-loops is not simple. To operate correctly the delay-locked-loop must be initialised with a known delay to ensure proper locking and sufficient range in both corrections. This requires extra circuitry.

The prior art infinite range analogue delay-locked-loop is a combination of several simple analog delay-locked-loops. Each has its own range limits and start up circuitry. The combination must be also be controlled, meaning higher order control loops, and therefore more complex stability issues. On the other hand the prior art digital or semi-digital delay-locked-loop has multiple taps from which the delay may be selected in steps. Since the step size is generally too coarse, the required phase may be generated by interpolation. This interpolation flattens the clock edges and therefore introduces jitter. This method also has complex control circuitry. The taps required on the delay-locked-loop delay line further limit its maximum frequency of operation.

A problem of the known delay-locked loops is that they not applicable over a very wide frequency range with high frequency while retaining good stability. Furthermore the known delay-locked-loops are not suitable for accurate receiver clock generation in systems in which the data rate is not exactly equal to the receiver reference clock rate.

It is an object of the invention to provide an improved delay locked loop which eliminates the above-mentioned disadvantages.

To this end, according to the invention, the delay-locked-loop comprises at least two delay elements, of which a first delay element has a positive delay line and an input for receiving a clock, and of which a second delay element has a negative delay line and an input for receiving a clock, a clock selector for selecting the clock from one of the two delay lines, a phase detector with an input for receiving data and for comparing the phase of the data to that of one of the clocks, and a control block which produces control signals for controlling the two delay lines such that they react in opposite directions.

The invention is based on the insight that a delay-locked-loop can use two parallel delay lines which react in opposite directions to a change in a control signal. According to the invention the delay-locked-loop is applicable over a very wide frequency range (typically higher than a prior art delay-locked-loop) while retaining good stability. The delay-locked-loop in accordance with the invention is also physically small and shows good noise and jitter immunity. The delay-locked-loop according to the invention may continuously increase or decrease the delay inserted into a clock path. This is equivalent to decreasing or increasing the frequency slightly. Therefore the delay-locked-loop according to the invention is suitable for accurate receiver clock generation in plesio-synchronous communication systems in which the data rate is not exactly equal to the receiver reference clock rate. Such systems have traditionally used power- and area-expensive over-sampling techniques in the receiver. With the invention not only is power area considerably reduced, but also the sampling accuracy is improved.

The delay locked loop is further characterized in that the control block interprets an output of the phase detector depending upon a currently selected delay line and compares the phase of the clocks out of the two delay lines.

In a further embodiment of the invention the control block determines a required delay line and indicates this to the clock selector, when the delay lines are aligned.

In a further embodiment of the invention a common control signal VC controls the two delay lines, which react in opposite directions to a change in the common control signal Furthermore the delay line is chosen such that the control signal VC required to produce a needed delay always stays within a desired range between a control signal level Vmin and a control signal level Vmax. This is for optimum performance, but not a requirement for the principle of operation.

A further advantage of the invention is that the control signal levels Vmin and Vmax are points at which the delays though the two delay lines are such that the clock may be selected from either delay lines.

The delay-locked-loop is further characterized in that if the control signal Vc increases, the delay through positive delay line increases and the delay through negative delay line decreases. According to claim 9 the phase detector produces a up signal which causes an increase in control signal Vc when the first delay line is selected and a decrease when the second delay line is selected.

The delay-locked-loop is further characterized in that the phase detector produces a down signal which causes an decrease in control signal Vc when the first delay line is selected and an increase when the second delay line is selected.

A feature of the invention is that the delay-locked-loop consists of a clock multiplexer which selects an output clock signal from one of two clock signal derived from the negative delay line and the positive delay line respectively.

In a further embodiment the delay-locked-loop consists of a delay line selector which tracks the delay line outputs.

A feature of the invention is that the delay line selector detects that the boundary of an operating area has been reached and selects the delay line so that the control signal VC is driven back into the operating area, when the clock signals derived from the negative delay line and the positive delay line are in phase.

According to the invention the delay line selector uses the current up and down signals from the data phase detector to decide which of the two clock signals becomes an active sampling clock.

In a preferred embodiment of the invention the delay-locked-loop is characterized in that at the boundary of the operating area it increases its delay when the control signal VC is driven towards a centre position Vmid, when the phase detector indicates that the sampling clock is too early. Furthermore the delay-locked-loop is characterized in that at the boundary of the operating area it decreases its delay when the control signal VC is driven towards a centre position Vmid, when the phase detector indicates that the sampling clock is too late.

Furthermore the delay-locked-loop is characterized in that the delay lines are arranged as a pair with opposite delay characteristics.

In a preferred embodiment of the invention the delay line selector chooses the delay line which requires the control signal VC to change towards the centre of the operating area when the delay-locked-loop reaches the boundary of the operating area. According to another embodiment of the invention the DLL may be initialised by forcing identical delays in each delay line.

In a preferred embodiment each delay line needs to control the delay over a range of only ¹A clock period (as shown in fig.3). Therefore there is no need to build a long delay chain to obtain a multi-clock period delay. Because the maximum delay in a delay line is small, the gain (delay per unit control level) is low, leading to improved stability and reduced sensitivity to noise on the control voltage (improved jitter performance) or from supplies. The power consumption is also low. These are all advantages that are desirable to retain even for lower frequency output clocks.

In a further embodiment of the invention each delay line may control the delay over a range of more than ¹A clock period.

In a further embodiment of the invention for lower frequency data rates, the delay lines may still be kept small if the clock applied to the clock input is a multiple of the required frequency. The clock is then frequency divided on the output.

For this embodiment the length of the up and down signals from the phase detector are matched to a rate of the input clock of the positive delay line and the negative delay line. A method of delaying in a clock path is object of claims 23 and 24.

A method of using the invention for locking a clock to a data stream with a slightly differ frequency is an object of claims 25 and 26.

The invention will be described in more detail with reference to the accompanying drawings, in which

Fig. 1 is a block diagram of a delay-locked-loop according to the invention

Fig. 2 is a another embodiment of a block diagram of a delay-locked- loop according to the invention

Fig. 3 is a delay characteristic in accordance with the invention

Fig. 4 illustrates a diagram of a delay inserted in delay lines in accordance with the invention

Fig. 5 shows a diagram of a control signal for constantly changing delay Fig. 6 shows a circuitry of a delay line according to the invention

Fig. 7 shows a circuit of delay lines according to Fig. 6 arranged as a pair of opposite characteristic in accordance to the invention Fig. 8 shows a circuit of delay line selector according to the invention

Fig. 9 shows a diagram of signals of a delay line selector according to the invention

Fig. 10 shows a clock alignment detection circuit according to the invention Fig. 11 shows a circuit of a delay line comparator according to the invention Fig. 12 illustrates a delay characteristic showing undesired operating areas in accordance of the invention Fig. 13 shows a block diagram with initialisation according of the invention

Fig. 1 shows a block diagram of a delay-locked-loop 100 according to one embodiment of the invention. The delay-locked-loop 100 comprises at least two delay elements, of which a first delay element has a positive delay line 10 and an input for receiving a clock, and of which a second delay element has a negative delay line 11 and an input for receiving a clock. Furthermore the delay-locked-loop comprises a clock selector 13 for selecting the clock from one of the two delay lines, and a phase detector 14 as well as a control block 12 which produce one or more signals for controlling the two delay lines 10,11. The phase detector 14 has an input for receiving data and for comparing the phase of the data to that of the selected clock, returning "phase up" and "phase down" pulses as appropriate. The control block 13 also interprets the output of the phase detector 14 depending upon the currently selected delay line, and adjusts the control signal(s) to correct the phase accordingly. Thus whether a "phase up" from the phase detector 14 results in an increase or a decrease of the control signal(s) depends upon the selected delay line. The control block 12 also compares the phase of the clocks out of two delay lines 10,11. When they are aligned, the control block 12 determines the required delay line and indicates this to the clock selector 13. The delay line 10,11 is chosen such that the control signal(s) required to produce the needed delay always stay within a desired range, of which the limits are the points at which the delays though the two delay lines 10,11 are such that the clock may be selected from either delay line. Because any required change of phase will always result in control signals that are within this range, the delay-locked-loop 100 according the invention can track an infinite change of phase in the input data. Provided that the rate of change of phase is within the bandwidth of the delay-locked-loop 100, the delay- locked-loop 100 will track the data without slipping cycles. Therefore, the invention may be used to produce a clock locked to a data stream that has a frequency slightly differing from that of the received clock.

For a detailed description and a better understanding of the invention the novel delay-locked-loop will be explained in conjunction with the fig. 2 to 13. This refers to an embodiment of the invention that uses a common control signal to control both delay lines. It is understood that it is possible to use separate control signals for the delay lines without changing the basic principle of operation.

DLP (positive delay line) 10 and DLN (negative delay line) 11 in fig. 2 and 3 are two delay lines. As mentioned above the only necessary difference between them is that they react in opposite directions to a change in the control signal. If control signal VC increases, the delay through delay line DLP 10 increases and the delay through delay line DLN 11 decreases. Fig. 3 shows the resulting delay characteristic of the two delay lines 10,11. The inverter on the output of DLP 10 in fig. 2 adds 180 degrees phase shift so that the delay-locked-loop 100 operates around its midpoint A in fig. 3. With differential clock signals, the inversion causes zero delay. With single ended clocks a small delay is introduced which shifts the operating point slightly away from midpoint A in fig. 3. This does not affect the operating principle. This inverter optimises the position of the operating area. It is not, however, inherently needed by the invention. As shown in fig. 4. the clocks clk n and clk_p are in phase when control signal Vc = Vmin and the clocks clk n and clk_p 180 degrees out of phase when the control signal Vc=Vmid. As Vc moves from Vmin to Vmax in fig. 3 clk_p is delayed by half a clock period and clk n is advanced by half a period. Thus the relative change is one period and clk n and clk_p are in phase again. Whenever clk n and clk_p are in phase they are indistinguishable so either of them may be selected as the source of the output clock clk m. The clock multiplexer 18 in fig. 2 is controlled by the delay line selector 20 and selects the output clock clk m from one of the two clocks clk n and clk_p derived from the delay lines DLP 10 and DLN 11 respectively. The phase detector 14 in fig. 2 compares the phase of clk m with that of the incoming data. An Alexander type phase detector 14 is typically suitable. Other types are possible. Cross- Mux 16 in fig. 2 cross-multiplexes the up and down signals from the data phase detector 14 in fig. 2 according to the currently selected delay line. Thus an up signal from the phase detector 14 in fig. 2 causes an increase in control signal VC when one delay line 10,11 is selected and a decrease when the other is selected, and vice versa for a down signal. The output may be forced by the delay line selector 20 in fig. 2 when the selected clock changes. The loop filter 15 in fig. 2 integrates the output of the Cross- Mux 16 in fig. 2 to produce the control signal VC for the delay lines 10,11. Typically this may be a charge pump and loop capacitor circuit.

The delay line selector 20 in fig. 2 tracks the delay line output. When the clk n and clk_p are in phase, the delay line selector 20 in fig. 2 detects that the boundary of the operating area 19, i. e. the box B-C-D-E in Fig. 4, has been reached and selects the delay line so that VC is driven back into the operating area 19, i.e. towards Vmid in fig. 3. The delay line selector 20 uses the current up and down signals from the data phase detector 14 in fig. 3 to decide which of the two clocks clk n and clk_p becomes the active sampling clock clk m. Should the data phase detector 14 indicate clk m is too early, then that clock is selected for which the corresponding delay line 10,11 increases its delay when the control signal VC is driven towards its centre position, Vmid. Should the data phase detector 14 indicate that clk m is too late, then that clock is selected for which the corresponding delay line decreases its delay when VC is driven towards Vmid. As an example assume that the data is arriving at a slightly lower rate than the clock. The phase detector 14 in fig.2 will constantly want to increase the delay in the delay line. Referring to fig. 3 and 5. the control signal will change as follows: C → E (switch to DLN 11) D → B (switch to DLP 10) C → E . Between points C — » E the loop behaves as a prior-art DLL using delay line DLP 10, and between points D -» B the loop behaves as a prior-art DLL using delay line DLN 11. At point E the delay line selector 20 in fig. 2 decides that delay line DLN 11 will bring the control signal back towards its centre point Vmin in fig. 3 since the phase detector 14 in fig. 2 indicates that more delay is required, so the system needs to increase delay while reducing the control signal level.

Two pieces of information are required - when the operating area 19 in fig. 3 is violated, and which delay line 10, 11 has the greater delay. As shown in fig. 3 the allowed (desired) limits of the control signal are Vmin and Vmax, which by definition are the points at which the delays though the two delay lines 10, 11 are such that the clock may be selected from either delay line. At this point the delay line selector 20 in fig. 2 chooses the delay line which, given the current phase of the data, requires the control signal to change towards the centre of its allowed range (towards Vmid in fig. 2).

The delay line selector 20 in fig. 2 consists of a phase detector to indicate when clk n and clk_p are in phase plus logic to determine whether the operating point is at BC (VC =Vmin) or DE (VC= Vmax) in fig. 3. A possible circuit of the operation is shown in fig. 8. At initialisation the system is brought to point A and the phase of the divided clocks are forced. Following this clk n and clk_p move relative to each other, thus affecting the signals p2_n2f and p2-n2r as indicated in fig. 9. Signal p2f_n2r therefore indicates that the system has crossed its operating boundaries and that the control signal VC needs to be returned towards Vmid. Signal p2f_n2r indicates which delay line currently introduces more delay, and which direction VC must be changed. Because the clocks generating the p2f_n2f and p2f_n2r signals have half the input clock rate, recognition of the crossing of the operating area boundary may be delayed one clock, increasing clock jitter at the switching points. To avoid this an extra clock alignment detection circuit may be used as shown in fig. 10. If the system is within the operating area 19 (signal p2f_n2f is high) in fig. 3 and the clocks are not aligned (signal elks aligned) low, the delay line selector 20 in fig. 2 takes no action-the currently selected delay line continues to be used. If however, p2f_n2f, is low or clks attgned is high, then the delay line selector 20 selects the clock source depending upon p2f_n2r and the output of the phase detector as below:

The control signal VC is changed as determined by p2f_n2r, overriding the output of the cross-multiplexer 16 in fig.2.

For the embodiment shown in fig. 7 which shows a circuit of delay lines (starved inverter delay lines as shown in fig. 6) arranged as a pair of opposite characteristics the signal p2f_n2r (which indicates which delay line currently introduces more delay) may be generated directly from the delay lines as shown in fig 11, since the delay is determined by the current in the delay line which in turn is set by control signal Vc. Because the difference in the delays is quite large at the switching point, errors due to mismatch are negligible. PMOS and NMOS match those used in delay lines shown in fig. 8. At initialisation the system must be brought to within the area B-C-D-E in fig. 4 and 13 for optimum performance. The areas 24 (A-F-G), 25(D-E-H-J), 26 (F- G-K-L) in fig. 12 are possible, but unwanted operating areas. In a preferred embodiment force delays in each delay line are identical. The initialisation is simple- simply force the two delay paths to have equal delay. This can only occur at point A in fig. 3. In fact the system can start at any point within the working range defined by the box BCDE in fig. 3, so inaccuracies due to mismatch are unimportant for initialisation. Assuming the delay line structure described in fig. 7, and using the delay line comparator in fig. 11, the input to the loop filter 15 in fig. 13 is fed from the signal "more delay in negative delay lines", so that the loop works to force equal delays in the two delay lines 10, 11. The system therefore moves towards operating point A in fig. 12. An initializer 21 in fig. 13 controls the source of the input to the loop filter 15 in fig. 13 until the loop settles at point A in fig. 12. At this point the output of the delay line comparator 23 in fig. 13 will be near the midpoint of the supply rails. The resulting complete system is shown in fig. 13.

REFERENCE NUMERALS

100 delay-locked-loop

10 negative delay line

11 positive delay line

12 control block

13 clock selector

14 phase detector

15 loop filter

16 cross mux

17 inverter

18 mux

19 operating area

20 delay line selector

21 initializer

22 init mux

23 delay line comparator

24 area

25 area

26 area

Claims

CLAIMS:

1. A delay-locked-loop (100) comprising at least two delay elements, of which a first delay element has a positive delay line (10) and an input for receiving a clock, and of which a second delay element has a negative delay line (11) and an input for receiving a clock; - a clock selector (13) for selecting the clock from one of the two delay lines (10, 11); a phase detector (14) with an input for receiving data and for comparing the phase of the data to that of one of the clocks; a control block (12) which produces one or more control signals for controlling the two delay lines (10, 11) such that they react in opposite directions to a signal from the phase detector.

2. A delay-locked-loop as claimed in claim 1, characterized in that a common control signal is used to control the two delay lines which react in opposite directions to a change in the control signal.

3. A delay-locked-loop as claimed in claim 1 or 2, characterized in that the control block (12) interprets an output of the phase detector (14) depending upon a currently selected delay line.

4. A delay-locked-loop as claimed in claims 1 to 3, characterized in that the control block (12) interprets an output of the phase detector (14), such output depending upon a currently selected delay line.

5. A delay-locked-loop as claimed in claims 1 to 4, characterized in that the control block (12) compares the phase of the clocks out of the two delay lines (10, 11).

6. A delay-locked-loop as claimed in claims 1 to 5, characterized in that the control block (12) determines a required delay line (10, 11) and indicates this to the clock selector (13).

7. A delay-locked-loop as claimed in claim 6 , characterized in that the delay line is chosen such that the control signal(s) required to produce the needed delay always stay within a desired range.

8. A delay-locked-loop as claimed in claim 7, characterized in that the maximum and minimum normal operating boundaries for the control signal levels are points at which the delays though the two delay lines (10, 11) are such that the clock may be selected from either delay line (10, 11).

9. A delay-locked-loop as claimed in claims 2 to 8, characterized in that if the con-trol signal VC increases, the delay through positive delay line (10) increases and the delay through negative delay (11) line decreases.

10. A delay-locked-loop as claimed in claim 2 to 9 characterized in that the phase detector (14) produces a up signal which causes an increase in control signal Vc when a first delay line is selected and a decrease when a second delay line is selected.

11. A delay-locked-loop as claimed in claims 2 to 10 characterized in that the phase detector (14) produces a down signal which causes an decrease in control signal VC when the first delay line is selected and an increase when the second delay line is selected.

12. A delay-locked-loop as claimed in claims 1 to 11 characterized in that the delay-locked-loop (100) consists of a clock multiplexer (18) which selects an output clock signal from one of two clock signals derived from the negative delay line (11) and the positive delay line (10) respectively.

13. A delay-locked-loop as claimed in claims 1 to 12 characterized in that the delay-locked-loop (100) consists of a delay line selector (20) which tracks the delay line outputs.

14. A delay-locked-loop as claimed in claim 13 characterized in that the delay line selector (20) detects that the boundary of an operating area (19) has been reached and selects the delay line so that the delay line control signal(s) is (are) driven back into the operating area (19), when the clock signals derived from the negative delay line (11) and the positive delay line (10) are in phase.

15. A delay-locked-loop as claimed in claims 13 or 14 characterized in that the delay line selector (20) uses the current up and down signals from the data phase detector (14) to decide, which of the two clock signals becomes an active output clock.

16. A delay-locked-loop as claimed in claim 15 characterized in that the delay through the delay-line increases when the control signal VC is driven towards a centre position Vmid, when the phase detector (14) indicates that the output clock is too early and when a boundary of the operating range is reached.

17. A delay-locked-loop as claimed in claim 15 characterized in that the delay through the delay-line decreases when the control signal VC is driven towards a centre position Vmid, when the phase detector (14) indicates that the output clock is too late and when a boundary of the operating range is reached.

18. A delay-locked-loop as claimed in claims 1 to 17 characterized in that the delay lines (10, 11) are arranged as a pair with opposite characteristics.

19. A delay-locked-loop as claimed in claims 13 to 16 characterized in that the delay line selector (20) chooses the delay line which requires the control signal VC to change towards the centre of the operating area (19).

20. A delay-locked-loop as claimed in claims 14 to 16 characterized in that force delays in each delay line are identical.

21. A delay-locked-loop as claimed in claim 20 characterized in that two delay path of the delay lines (10, 11) have equal delay.

22. A delay-locked-loop as claimed in claims 1 to 21 characterized in that the length of the up and down signals from the phase detector 14 are matched to a rate of the input clock of the positive delay line (10) and the negative delay line (11).

23. A method of delaying in a clock path, the method comprising: receiving a clock into two delay elements, of which a first delay element has a positive delay line (10) and an input for receiving a clock, and of which a second delay element has a negative delay line (11) and an input for receiving a clock; selecting the clock from one of the two delay lines (10, 11); receiving data and for comparing the phase of the data to that of the selected clock; - controlling the two delay lines (10, 11) which react in opposite directions to a change in the control signal.

24. A method of delaying in a clock path, the method comprising: receiving a clock into two delay elements, of which a first delay element has a positive delay line (10) and an input for receiving a clock, and of which a second delay element has a negative delay line (11) and an input for receiving a clock; selecting the clock from one of the two delay lines (10, 11); receiving data and for comparing the phase of the data to that of the selected clock; controlling the two delay lines (10, 11) such that they react in opposite directions.

25. A method as described in claim 23 or 24 in which the received clock has a frequency near to but not necessarily equal to the data rate, such that the difference in frequency causes a continuously changing delay in the clock path.

26. A system as described in claim 25 in which the maximum difference in frequency between the received clock and the data is defined by the bandwidth of the delay-locked-loop.