GB2373143A

GB2373143A - Synchronising base stations

Info

Publication number: GB2373143A
Application number: GB0200698A
Authority: GB
Inventors: Anthony Peter Hulbert
Original assignee: Roke Manor Research Ltd
Current assignee: Roke Manor Research Ltd
Priority date: 2001-01-17
Filing date: 2002-01-14
Publication date: 2002-09-11
Also published as: GB0101200D0; GB0200698D0

Abstract

A communications network comprises a radio network controller (RNC) and a number of base stations. A system for and method of synchronising base stations, one of the base stations has the correct time. The out-of-synchronisation base stations are instructed to search for frequency acquisition transmissions, and base station having the correct time is instructed to make a continuous frequency acquisition transmission, the out-of-synchronisation base stations acquire the time and frequency of the frequency acquisition transmission. The continuous frequency acquisition may be completed over a time period of 0.7 to 50 ms, and all base stations are synchronised in less than one second. When an out-of-synchronisation base station has acquired the frequency from the base station with the correct time, the now-in-synchronisation base station preferably signals this to the RNC, and produces its own frequency acquisition transmissions. Frequency shift acquisition is performed in a first course step followed by a fine tuning step using an automatic frequency control loop.

Description

Method of synchronising base station transmitters. This application relates to a method of frequency acquisition of base stations of a communication system so that they acquire the frequency of a master base station (Node B frequency acquisition) and its use in synchronising base station transmitters A significant cost element in mobile telephone base station (Node B) hardware is the reference oscillator. One of the NodeBs (or cells) in an radio network system will contain a GPS receiver, but it is prohibitive to include a GPS receiver in every base station.

Over-the-air NodeB synchronisation has been proposed to remove the need for GPS receivers in other NodeBs in order to save cost. In other words one master base station can ensure other base stations are synchronised to it by sending out synchronisation signals.

However, expensive ovened crystal oscillators are required to obtain the necessary absolute reference frequency oscillator accuracy, which is a prerequisite of synchronisation of time.

It is therefore an object of the invention to introduce a method whereby NodeBs with relatively poor absolute accuracy can acquire the frequency of a Node B having an accurate clock; e. g. a Node B containing the GPS receiver. Once suitably trained, these NodeBs can act as over-the-air frequency references for other Node B's which are in range of them but which are out range of the Node B containing the GPS receiver, and so on until all oscillators are trained.

Known Node B synchronisation is based on the transmission and reception of Node B sync bursts. In order to maximise the range/reliability of synchronisation links between NodeBs, the sync burst has been defined as

a long code, occupying most of a complete time slot. Whilst advantageous in terms of received sensitivity, the maximised code length leads to an increased sensitivity to frequency errors. For a code length of 2048 chips at 3.84 Mcps, a frequency error of 1875 Hz would result in a complete cycle of phase shift over the code period, further resulting in zero output from a correlator. For a 2 GHz transmission this corresponds to an error of approximately 1 ppm. In order to minimise costs it is desirable to operate with oscillators with frequency errors up to : 3 ppm. This would not be possible without searching in frequency.

The inventor has determined that introducing an optional frequency acquisition phase as part of the NodeB Sync procedure but preceding the synchronisation operations which are currently standard. Furthermore under the invention one or more NodeBs, initially just the NodeB containing the GPS receiver, transmit continuously.

The invention therefore comprises, in a communication network comprising a radio network controller (RNC) and a number of base stations, a method of synchronising base stations of, one of which has the correct time comprising the steps of ; a) instructing the out-of-sync base stations to search for frequency acquisition transmissions ; b) instructing the base station with the correct time to make a continuous frequency acquisition transmission ; c) allowing one or more out of sync base stations to acquire time and frequency of frequency acquisition transmission.

Preferably the period of time for the continuous frequency acquisition transmission is less than 5ms but more that a time slot (typically 667ass second).

Given that commissioning a new Radio Network System (RNS) will be a very infrequent event which can be arranged for a time of low traffic activity (e. g. during the night), interference with existing RNSs using the same frequency can be neglected. The use of different NodeB sync burst codes for different RNSs would prevent longer term disruption of the existing RNS from these transmissions.

The normal NodeB sync code is repeated in every time slot for the frequency acquisition burst. The code structure currently proposed for NodeB sync involves a so-called complementary extended code (CEC).

This entails the concatenation of two codes complementary codes, each with cylic extension wherein a section of the beginning of the code is repeated at the end. Complementary codes are code whose even autocorrelation functions are the inversion of each other apart from at the aligned position where they are the same. The cyclic extension length can be increased to make the transmission continuous.

Figure 1 contrasts the standard known synchronisation time slot allocation with that according to the invention for the additional phase of frequency acquisition. In normal TDMA there are 15 time slots 2 per frame 1. Figure la shows a number of frames 1. One slot in every frame, the Physical Random Access Channel or PRACH, is utilised for the synchronisation signal, and lasts typically 667 microseconds, i. e. a 1 5tu of the total frame time of 10ms. Figure 1 b shows how the step of frequency acquisition according to the invention uses the timeslots in the frames. As can be seen the frequency acquisition signal is continuous and in each time slot for a

number of frames contains frequency acquisition data. It is to be noted that there is usually a small guard period in each time-slot and the term "continuous"meaning that consecutive time slot are used for frequency aquisition.

Any attempt to acquire frequency using the figure la method, i. e. a short slot duration, is very difficult because the base station has to search not only the correct frequency but has to know the right time in which to do it.

The method according to the invention overcomes this problem.

The following shows examples only of preferred methods which fall within the scope of the invention may be embodied.

In a preferred embodiment, two phases for frequency acquisition at the NodeB/cell level occur :1. In the first phase, matched filters to tile two complementary codes are

run ; suitable matched filters can use the structure described in the paper by S. Z. Budisin,"Efficient pulse compressor for Golay complementary sequences", Electronics Letters, Vol. 27, No. 3, pp. 219220, Jan. 1991, and are very efficient for the present application, requiring only 40 additions/subtractions per sample to perform matched filtering over both codes in parallel. Over a period of 1.5 slots it is possible to guarantee to receive both codes in either order.

However, if the frequency error is large then neither code may be detected. It is advantageous therefore to frequency shift the received signal prior to correlation. If the codes are not detected during a period of 1.5 slots, the frequency is shifted and the process repeated. If a maximum loss of 1 dB is allowed during each code due to frequency error, the maximum permissible frequency error is : 0. 26 cycles over 1024 chips or about 1 kHz corresponding to 0. 5 ppm at 2 GHz. Thus

the frequency can be stepped by 1 ppm in each interval. If, e. g., the search is over : 3 ppm then this corresponds to 6 steps or 9 slot periods. If the step in analogue using the temperature controlled, voltage controlled crystal oscillator or-the step size can be subject to the uncertainty of the oscillator gain constant. Suppose the gain accuracy is : f : 20%. If a maximum step size of 1 ppm is required (for oscillator at the +20% end of operation) then the nominal step size will need to be 1/ (1+0. 2) = 0. 833 ppm. The minimum step size will then be (1-0.2) x 0.833 ppm = 0. 666 ppm. In order to guarantee covering the whole range of : f : 3 ppm with the minimum step size we will need 9 steps or 14.5 slot periods. Thus, with relatively simple hardware it should be possible to search the whole frequency range in less than 15 slots, i. e. less than one frame.

2. Once first phase (coarse) frequency acquisition has been achieved, the second phase acquisition is invoked. This can be done by using an automatic frequency control (AFC) loop. The frequency error detector for this can be implemented by correlating against the second code and measuring the phase difference between this and the first code.

Alternatively, the correlation scores for the two codes can be added together and compared with the phase of the sum of the scores for the previous two codes. The latter approach has a smaller maximum frequency error range but improved noise performance.

Suppose the AFC loop parameters are set for maximum rms error of 10 ppb. This corresponds to 20 Hz at 2 GHz or 0.01 cycles over 2048 chips. Simulations show that a suitable AFC loop can settle to well within 10 ppb in under 30 iterations (time slots) for a total (both

complementary codes) despread output signal to noise ratio of only 6 dB. Thirty time slots is equivalent to 2 frames (20 ms) Thus a single NodeB/cell should be able to acquire frequency in only 30 ms. On this basis, all of the neighbours of the GPS NodeB/cell should be in sync within 30 ms. Their neighbours should have acquired frequency within a further 30 ms. The total time for all NodeB/cells in an RNS will be approximately M/2 X 30 ms on average where M is the number of 'hops'across the network. The factor of 2 arises because the GPS NodeB/cell will typically be in the centre of the network with frequency acquisition fanning out in all directions. Typically, for a two dimensional deployment, the number of hops will grow as the square root of the number of NodeB/cells. Thus, even for a large RNS containing of the order of 1,000 NodeB/cells, the time to acquire frequency should not exceed one second. This is important for two reasons: it keeps the continuous transmission time down; and it minimises the period over which the oscillators can drift As stated earlier, once an individual NodeB/cell has achieved frequency acquisition it starts to transmit. In order to avoid cross code interference it is necessary that the new transmitter sets its transmit timing close to the previously received timing. Although it can no longer receive, its timing should not drift far in time since its clock frequency is now close to that of others that have transmitted. For example, an error of 50 ppb would hold timing to within 50 ns ( < 1/4 chip) over a second. Any NodeB/cell (s) receiving from this and other transmitters will typically resolve several paths from each and resolve the paths from the different transmitters according to their times of arrival. This provides an element of diversity. It also allows the measured frequency to be formed as some average over the

frequencies of the signals received. This should mean that standard deviation of the frequency error of the composite signal being received should be lower than that of any individual signal. This should mitigate any build up in frequency error across the network when NodeB/cells increasing distant (in terms of hops) from the GPS NodeB/cell acquire frequency.

All neighbours to the GPS NodeB/cell will have frequency error taken as independent samples from a Normal distribution with standard deviation (say) 10 ppb, introduced by the contribution from receiver noise in the AFC loop. Their neighbours will see between one and three transmitters closer to the GPS NodeB/cell depending on their position as illustrated inFigure 2The NodeB/cells along the diagonal will all only see one neighbour nearer to the GPS NodeB/cell. In this case the variance grows linearly with the distance (in hop counts) from the the GPS NodeB/cell.

Thus, for example, after 4 hops the standard deviation would have doubled to 20 ppb. Where there are multiple paths to a NodeB/cell the variance still grows linearly with the distance but this growth is scaled downwards by the average number of paths. For example, with 3 paths, the standard deviation after 4 hops would be about 14 ppb.

Once frequency acquisition has been achieved, the normal (previously defined) NodeB sync scheme can be run.

Once full NodeB sync has been established the reference frequency accuracy can be maintained by any combination of the following two operations :

1. Adjust frequency in the direction which tends to compensate for the need to perform timing updates 2. Compute small frequency adjustments based on frequency error measurements from individual NodeB sync messages. In tins case the measurements can be performed by comparing the phase of the correlator output for the second complementary code to the phase of the correlator output for the first complementary code. These measurements are somewhat restricted in performance because they are infrequent. This obviously reduces the update rate. In performing frequency updates in this way it is important only to respond to measurements of signals from NodeB/cells which are closer to the GPS NodeB/cell than the receiver.

Frequency acquisition for NodeB/cells which are late entrants to the system.

With the above described simultaneous transmission of sync bursts in a given area, a late entrant should hear a sync burst from at least one of its neighbours every 2.56 seconds. With nine serial searching steps for frequency as described earlier, this would mean a maximum period of 9 x 2.56 = 23 seconds to detect a sync burst on the correct frequency, and this is acceptable. Even if the gap between sync bursts is extended, the period would still only be of the order of minutes. The initial detection would reduce the frequency error to : 0. 5 ppm. From here some frequency correction could be applied from the individual sync bursts but the main method will be based on timing updates as for the normal case.

Frequency Error Detector The ideal means for achieving phase and frequency lock is not a frequency detector but a phase detector which is able to provide an unambiguous phase measurement over an unlimited range of input phase-ie not limited to ? t radians. It is easy to see that, within a closed loop circuit, a frequency discriminator followed by an integrator is in fact a phase detector with extended phase response. However the general frequency detector, in this mode tends to lead to high levels of noise. These effects are greatly alleviated by forming a linearised frequency detector in which some of the noise between adjacent samples is anticorrelated and therefore removed. Suppose the input to the phase detector consists of samples, yn, where yn = sn + Nn and sn is the nth received signal component, subject to some phase rotation, and Nn the corresponding sample of complex AWGN.

The ideal phase detector simply applies the'argument'function to the successive yn values. We can also obtain the phase difference between adjacent samples by fonriing z,, Note that we have -1} =--, If we now feed these phase differences to an accumulator, we form

n2 E n Yn-l} From (1) we can see we have y) = zv,,,-zx, -1. n= =n,

This means that the noise on the measurements in intermediate stages of the summation contributes nothing to the overall results. In effect, a component of the noise on Y y I is anticorrelated with a component of n the noise on z {, }

Claims

Claims 1. In a communication network comprising a radio network controller (RNC) and a number of base stations, a method of synchronising base stations of, one of which has the correct time comprising the steps of a) instructing the out-of-synchronisation base stations to search for frequency acquisition transmissions; b) instructing the base station with the correct time to make a continuous frequency acquisition transmission ; c) allowing one or more base stations to acquire time and frequency of frequency acquisition transmission.
2. A method as claimed in claim 1 wherein said continuous frequency transmission acquisition is done over a time of period of between 0. 7ms and 50ms.
3. A method as claimed in claims 1 or 2 wherein all nodes B's are synchronised in less than one second.
4. A method as claimed in claim 1 wherein when a previously out of synchronisation base station has acquired frequency form the base station with the correct time, the additional step of a) signalling this to the radio network controller b) and starts to produce its own frequency acquisition transmissions.
5. A method as claimed in any preceding claim wherein the frequency shift acquisition is performed in a first coarse step followed by a fine tuning step using an automatic frequency control loop.
6. A communication network comprising a radio network controller (RNC) and a number of base stations, and having means to synchronise base stations, one of which can be set to have the correct time comprising : a) means to instruct the out-of-synchronisation base stations to search for frequency acquisition transmissions; b) means to instruct the base station with the correct time to make a continuous frequency acquisition transmission; c) means to allow one or more base stations to acquire time and frequency of frequency acquisition transmission.
7. A network as claimed in claim 6 wherein said continuous frequency transmission acquisition is done over a time of period of between 0. 7ms and 50ms.
8. A network as claimed in claims 6 or 7 wherein all nodes B's are synchronised in less than one second.
9. A network as claimed in claim 6 wherein when a previously out of synchronisation base station has acquired frequency form the base station with the correct time, the additional step of a) signalling this to the radio network controller b) and starts to produce its own frequency acquisition transmissions.
10. A network as claimed in any claim 6 to 9 wherein the frequency shift acquisition is performed in a first coarse step followed by a fine tuning step using an automatic frequency control loop.