COMMUNICATION SYSTEM AND METHOD USING TRANSMIT DIVERSITY
FIELD OF THE INVENTION
The invention relates to a communication system .using transmit diversity, and to a method to be performed in such a communication system.
BACKGROUND OF THE INVENTION
Transmit diversity techniques provide advantageous solutions for increasing downlink capacity in 3G (Third Generation) systems within low-mobility environments. The complexity to implement transmit diversity mainly burdens the base station making the technique more suitable for low-cost handsets than, e.g., receive diversity. Open-loop and closed-loop transmit diversity techniques have already been standardized and improvements are being developed with 3GPP WCDMA FDD and TDD modes. Transmit diversity is considered with EDGE standardization as well. (Abbreviations: 3GPP, Third Generation Partnership Project; WCDMA, Wideband Code Division Multiple Access; FDD, Frequency Division Duplex using two distinct frequencies for uplink and downlink; TDD, Time Division Duplex using two distinct time slots on the same frequency for uplink and downlink.)
If the transmitter wants to utilize the state of the channel in FDD systems the receiver has to provide necessary information to the transmitter through some feedback mechanism. In general, when the correlation of the transmit antennas is small, the receiver has to determine and quantize
the eigenvector corresponding to the maximum eigenvalue of the channel correlation matrix which quickly becomes a cumbersome task when the number of transmit antennas is increasing .
One example of transmit diversity is described in: "A simple transmit diversity technique for wireless communications", S. M. Alamouti, IEEE Journal of Selected Areas of Communications, Vol. 16, No. 8, October 1998, pp. 1451-1458.
One of the problems of the prior art is to find optimal solutions for determining transmit weights when the number of possible weight vectors is large.
SUMMARY OF THE INVENTION
The invention provides a communication system and/or method as defined in any one of the attached claims, having an improved functionality.
Furthermore, the invention provides a network element (e.g. a user equipment such as a mobile station) having a function to calculate feedback bits to be sent the assigned transmission station of the Radio Access Network (RAN) such as RNC or BSS for improved reception e.g. by the user equipment.
The invention proposes several ways to arrive at good and effective solutions for determining e.g. transmit phases and/or transmit weights.
Generally, the invention provides a solution for increasing the downlink capacity e.g. in a wireless communication system in which the transmitter can be provided with only a limited information about the channel state.
In a preferred example, when only the total gain of some group of transmit antennas is needed it is proposed, according to one aspect of the invention, to send the same pilot sequence from the antenna group where the pilot sequences across the antennas are delayed. This scheme is applicable to other transmit diversity techniques as--well. For example, in a case where four transmit antennas are divided into two groups it is enough to send only two different pilot sequences. Simulation results without pulse shaping and sampling at the chip rate indicate that with ITU Pedestrian A channel model the correlation of selecting the best pair of transmit antennas is over 90% when compared with the scheme employing four different pilot signals, and with ITU Vehicular A, the correlation is still slightly more than 80%.
Thus, it is possible to apply some space-time coding scheme to the selected antenna pair and estimate the channel from the dedicated pilot channel. The four-antenna transmit diversity system utilizes only two different common pilot signals and therefore it is compatible with two-antenna transmit diversity schemes.
In accordance with preferred embodiments of the present invention, a two-fold target can be achieved: to facilitate efficient calculation of feedback weights in user equipment, and still providing a significant improvement in the received signal-to-noise ratio. For this, several new closed-loop algorithms are presented.
Furthermore, the disclosed systems and methods do not necessarily require reference signals to be sent from all the antennas simultaneously so that the number of required pilot
signals as well as the amount of signal processing in mobile terminal can be reduced.
The invention can be combined with progressive refinement of transmit weight, such as specified in the current 3GPP WCDMA FDD transmit diversity mode 1, as well as with Bayesian estimation of transmit weights in the base station.
Furthermore, the invention can be combined with predictive channel estimation in mobile terminal.
In accordance with preferred embodiments of the invention, the downlink capacity can be increased in a wireless communication system where the transmitter can be provided with only a limited information about the channel state.
Generally, the invention preferably relates to a radio communication link susceptible for fading, when multiple transmit antennas are available, and discloses, among others, a method to distribute the information in feedback bits in an as effective way as possible between multiple antennas. The limited number of feedback bits is used in an as effective way as possible to increase performance.
The invention provides, among others, a method to calculate feedback weights in FDD TX diversity with more than 2 tx elements, and is e.g. applicable in future 3G WCDMA releases.
BRIEF DESCRIPTION OF THE DRAWINGS
Figure 1 shows a block diagram of an embodiment of a receiving device which can be used in an embodiment of a communication system in accordance with the invention; and
Figure 2 illustrates an embodiment of a transmitting device which can be used in an embodiment of a communication system in accordance with the invention.
DETAILED DESCRIPTION O.F- PREFERRED EMBODIMENTS OF. THE
INVENTION
Figure 1 shows a block diagram of an embodiment of a receiving device which can be used in an embodiment of a communication system in accordance with the invention. The receiving device may be a user end terminal such as a mobile station/phone, and comprises a single receiving antenna 1. The received antenna signals are supplied to an RF receiver 2 which may e.g. perform a frequency conversion to IF frequency. The output of RF receiver 2 is supplied to a detection means 3 for signal detection, and parallely to a channel estimation means 7 carrying-out a channel estimation. The output 4 of the detection means 3 is supplied to the usual receiver components (not shown) for signalling and user traffic evaluation.
The channel estimation means 7 outputs the result of channel estimation to the detection means 3 as well as to a feedback calculation means 6 for calculating feedback bits or words, i.e. transmit weights, using algorithms corresponding to phase rotations and/or power adjustment. The calculated feedback bits or words are sent to the transmitting device (see Fig. 2) via a transmit antenna 5 of the receiving device.
Figure 2 illustrates an embodiment of the transmitting device in accordance with the invention. The traffic and control
channels used in the transmitting device are represented by an arrow 21 and are supplied to a multiplexer 22. The output of the multiplexer 22 is connected to a scrambling means23 for scrambling the output signals of the multiplexer 22 in a known manner. The scrambled signals output from scrambling device 23 are supplied to a spreading means 24 for performing a known spreading process. The spreaded output signals of spreading means 24 are supplied to inputs of a weighting means 25 having a plurality of paths each comprising an input and an output of the weighting means 25. Each output of the weighting means 25 is connected to an associated one of a plurality of antennas 211 to 27M via an interconnected RF transmision amplifier 2 <E>\ to 26M. The antennas 27χ to 27M are antennas both for transmission and reception of signals to and from receiving device shown in Fig. 1 and similar receiving devices e.g. of other users.
Further, a feedback decoding and weight calculating means 28 is provided having its inputs connected to the antennas 27ι to 27M and controlling, via its outputs connected to the weighting means 25, the weights apportioned to the paths of the weighting means and thus to the spreaded output signals of spreading means 24 before supplying these signals to the RF amplifiers 26ι to 2βM. The feedback decoding and weight calculating means 28 selects the feedback commands (feedback bits or words) calculated in and returned from the receiving device such as shown in Fig. 1, e.g. from mobile stations (MSs) .
In the following, at first two basic prior art ways will be briefly described, followed by a detailed description of the functioning of methods and devices in accordance with the invention.
There are basically two straightforward prior art ways to calculate the necessary feedback information describing the channel state:
1) Determine the maximum eigenvector of the channel covariance matrix and map it to the closest feedback vector.
2) Try out all 2N possibilities of rotations/gain adjustments determined by the N-bit feedback (FB) word and choose the one providing the largest SNR (signal-to-noise ratio) improvement or the largest SNIR (signal-to-noise-and-interference ratio) improvement.
3) Calculate feedback vectors independently on antenna basis. Assuming that the number of feedback per antenna is constant this requires M*2Λ (N/M) comparisons. Thus, the complexity of the algorithm is smaller than those of the algorithms 1) and 2) above, but at the same time the SNR improvement is degraded as well.
Contrary thereto, the invention proposes new algorithms for calculating feedback weights so as to quickly and easily control the transmission for good and strong reception by the receiving device such as MSs. Given the hierarchical structure of the algorithms used in the invention they provide an efficient way to determine the feedback word without eigenvector calculations irrespective of the number of transmit antennas or ultipath components. Furthermore, if the number of feedback (FB) bits/antenna is kept constant then the number of comparisons to determine the feedback word does not increase exponentially with respect to the number of antennas but only linearly as will be described below in greater detail.
Thus, the complexity of the new algorithms is similar to case 3) above while the SNR improvement is similar to cases 1) and 2) .
In case of limited feedback capacity, typical to FDD system, it is possible to transmit a common pilot signal, not from every antenna, but from every antenna group only, reducing the number of required pilot sequences, reducing the amount of signal processing in the mobile station, and increasing the accuracy of the measurements. In the case of an example of four Tx (transmit) antennas and below described algorithms 2 and 3 corresponding to phase rotations, it is sufficient to send simultaneously only two common pilot signals instead of four. On the other hand, it is also possible to send simultaneously a different pilot signal from each antenna so that the antenna groups can be arbitrarily calculated in the receiver. The number of simultaneous pilot signals can therefore be reduced. Further, the methods and systems according to the invention can inherently be power balanced so that all antennas transmit with the same power.
When channel is changing slowly, the accuracy of the feedback word can be easily improved by a preferred implementation of the invention due to the hierarchical structure, and the same algorithm can be used with different feedback modes if necessary. Further, the inter-group phase adjustment achievable according to the invention has a large effect onto the SNR improvement as can be seen in Table 1 (see below) .
Further, the hierarchical structure of the new algorithms provides a simple and efficient way to apply feedback bit words of length 3, 4, 5, ... This feature is not available in general if grouping is not applied.
First, feedback algorithms corresponding to phase rotations will be described below.
Three classes of feedback algorithms for calculating possible phase rotations will be given in this section. In the following, at first simple examples of each of these algorithms are presented and thereafter general -formulations for all three classes of algorithms are given.
Since complex channel parameters can be viewed also as vectors in L-dimensional complex space (here L is the number of channel paths) these parameters will subsequently be termed as "vectors" or "signal vectors" and their "length" is given by
For example, a statement "vector 2 is rotated by 180 degrees" means that transmitter changes the phase of the signal transmitted from antenna 2 by 180 degrees.
The summing-up of the vectors and the check whether or not the summing of the vectors is destructive, i.e. the resultant sum vector has a smaller amplitude (length) than at least one of the original vectors, is performed in the channel estimation and feedback calculation means 7, 6 shown in Fig. 1. The calculation of feedback bits to be sent to means 28 is performed in the feedback calculation means 6. The vector rotation is effected in the transmitting device, e.g. by setting the polarity of the weights or by otherwise changing the phase of the signals sent from the antennas 27x to 27M.
As an example case, a system will be considered where four transmit antennas and one receive antenna is available. Assume that the feedback word consists of 3 feedback information bits concerning the phases of channel parameters. The vectors are designated as "z" with an index corresponding to their number.
A signal vector can be associated to the corresponding transmit antenna, or alternatively, a signal vector may correspond to a beam transmitted from an antenna array.
First algorithm (3 FB bits) . First vector z__ is left untouched. If first and second vectors zi and z2 sum up destructively then the second vector z2 is rotated by 180 degrees. If the sum of first two vectors Zi and z2 and the third vector z3 sum up destructively then the third vector is rotated by 180 degrees. If the sum of first three vectors and the fourth vector z sum up destructively then the fourth vector is rotated by 180 degrees. That is, in the first stage, the second vector is weighted by w2 , if necessary, in such a way that the length of the sum z] A- w2z2 attains its maximum:
+ w2z2|| = max||z1 + w2z2| : w2 e {- 1,1}}
(1 bit used) . In the second stage, the third vector is weighted by w3 , if necessary, in such a way that
||z, + w2z2 + w3z3|| = max|z, + w2z2 + w3z3| : w3 e {-l,l}
( 1 bit used) .
In the last stage, the fourth vector is weighted by w4 , if necessary, in such a way that
z, + w2z2 + w3z3 + w4z4|| = max|z] + w2z2 + w3z3 + vf4z4|| : w4 e {-l,l}
Second algorithm (3 FB bits) . In the first stage, if the first and second vectors sum up destructively then the second vector is rotated by 180 degrees and if the third and fourth vectors sum up destructively then the fourth vector is rotated by 180 degrees. In the second stage, if the sum of first and second vectors sums up destructively with the sum of third and fouth vectors then third and fourth vectors are rotated by 180 degrees. That is, in the first stage, the second and fourth vectors are weighted byw2and w4 , if necessary, in such a way that
]z, + w2z2 J = max z, + w2z2| : w, e {-l,l}} ,
||z3 + w4z4|| = maxi|z, + w4z4|| : w4 e {- l,l}
(2 bits used) . In the second stage, the third and fourth vectors are weighted by w, , if necessary, in such a way that
||z, A- w2z2 + w3(z3 + w4z4)| = max|]z, + w2z2 + w3 (z3 + w4z4)| : w3 e {-l,l}
( 1 bit used) .
Third algorithm (3 FB bits) . In the first stage, if the sum of first and second vectors sums up destructively with the sum of third and fourth vectors then third and fourth vectors are rotated by 180 degrees. In the second stage, if the sum of second and third vectors sums up destructively with the sum of first and fourth vectors then first and fourth vectors are rotated by 180 degrees. In the third stage, if the sum of first and third vectors sums up destructively with the sum of second and fourth vectors then second and fourth vectors are rotated by 180 degrees.
That is, in the first stage, the third and fourth vectors are weighted by w, , if necessary, in such a way that
||z, + z, + w, (z3 + z4)|| = maxfz, + z2 + w, (z3 + z4)|| : w, e {-l,l} }
(1 bits used) . In the second stage, the first and fourth vectors are weighted by w2 , if necessary, in such a way that
||z2 + w, z3 + w2 (z, + w, z4 )|| = max||z2 + wlz3 + w2(zl + w, 4 )|| : w2 e {- 1,1 }
(1 bit used) . In the third stage, the second and fourth vectors are weighted by w3 , if necessary, in such a way that
Iw
2z, + w,z
3 + w
3 (z
2 + w
2 w,z
4 )|| =
+ w,z
3 + w
3 (z
2 + w
2 w,z
4 )|| : w
3 e {- 1
(1 bit used) .
Description of General Algorithms
Assumptions for the first algorithm: M transmit antennas, N*( -1) feedback bits available.
First algorithm. First vector is left untouched. Second vector is weighted by w2 , if necessary, in such a way that
+ w
2z
2|| = maxjz, +
e
/2,r("-'
)/2' , n e {l,2,...,2
w }
This can be done by using the information content of first N feedback bits. In the second stage the third vector is weighted by w3 , if necessary, in such a way that
z, + w
2z
2 +
= ma lz, + w
2z
2 + w
3z
3| : w
3 =
e j2π{-
n~ )l'1'
' , n e
}
A general step of this algorithm is such that the k'th vector is weighted by wk , if necessary, in such a way that
Since M-l steps is implemented in total, we need N*(M-1) feedback bits. The order of the vectors may also depend on some ranking scheme. For example, in the first stage
, + w
22
2| = max j|z, + w
2z
2| : w
2 =
e j2π{n~l)'
2" ,n e
} ,
where z'2 =argmax(|l z2 ||, ... ,|| zM ||) or z = argmax(|| z, + z *2 \\, ... ,|| z, + z *M ||) where in the last case z *2 , ... , z *M refer to vectors with "the best rotation" with N bits.
Computational complexity of the first algorithm: In all stages it is necessary to compute only 2H different values
and make 2N -1 comparisons. In total we need to compute (M -\) - 2N different numeric values and make (M - 1) - (2N -1) comparisons. If we would like to go through all alternatives (and seek optimal solution) then we would need to compute 2(Λ )Wdifferent values and make 2(M_1)A' -1 comparisons.
Thus the amount of work in the case of the presented algorithm grows linearly with the number of antennas whereas in the general case (goal is the optimal solution) the amount of work grows exponentially with the number of antennas. (It is assumed here that the number of FB bits/antenna is kept constant)
Assumptions for the second algorithm: M transmit antennas, the number of feedback bits is given in the description of the algorithm.
Second algorithm (multilevel algorithm) . In the first stage we divide all antennas into K groups of Mk , k = 1,2, ... , K, antennas. First algorithm is applied inside each of these groups. The number of needed feedback bits is
*,=ι
where Mk is the number of vectors in group k . In practise it is reasonable to choose M = M2 = ... = MK and N, = N2 =... = Nκ if possible. After the first stage we can view each group of Mk vectors as a single (sum) vector. Thus, in the second stage we have a set of K] (sum) vectors to be adjusted. Now we form R~ 2 new groups from R", (sum) vectors. Again, the first
algorithm is applied inside these new K2 groups. The number of needed feedback bits in this stage is
^2=Σ(^2-D-^2 •
The algorithm proceeds from this on in a similar manner. If we finish after L stages we find that the number of feedback bits in total is
/. K,
^ = ∑Nl = ∑∑(Mkι -l) - Nk l=\ /=! k,=\
Computational complexity of the second algorithm: Here the amount of needed work can be deduced similarly as in the case of the first algorithm. Since the obtained formulae are relatively complex it is just summarized that the amount of work in the case of the second algorithm grows linearly with the number of antennas whereas in the general case (goal is the optimal solution) the amount of work grows exponentially with the number of antennas. (It is assumed here that the " number of FB bits/antenna is kept constant) .
Example (4 Tx antennas, 4 FB bits) . First and third vectors are left untouched in the first stage. Second and fourth vectors are weighted by w2 and w4 , if necessary, in such a way that
||z, + w2z2|| = max|z, + w2z2|| : w2 e {- 1,1 } } , IK + ^A Z A \\ = maxf z3 + w4z4|| : w4 e {-l,l}
(2 bits needed) . In the second stage the sum vector z] + w2z2
is left untouched and we weight the sum vector z3+vi>4z4 by w3 , if necessary, in such a way that
z, + w2z2 + w3 (z3 + w4z4 )|| = maxjz, + w2z2 + w3 (z3 + w4z4 )|| : w3 e {1,7,-1,-7} }
( 2 bits needed)
Feedback rate corresponding to different stages of the algorithm can be different. For example, suppose that the correlation within the pairs (zl,z2) and (z3,z4) is stronger than between the pairs. This could happen when z are associated to antennas in an antenna array where spatial separation of the antenna pairs is larger than the spatial separation within the pairs. Alternatively, antenna pairs may have different polarizations. Then it is advantageous to calculate and send the feedback of the second stage of the algorithm more often than that of the first stage.
Example (8 Tx antennas, 11 FB bits). Vectors 1,3,5 and 7 are left untouched in the first stage. Vectors 2,4,6 and 8 are weighted by w2 , w4 , w6 and wgin such a way that
||z, +.w2z2|| = maxjz, + w2z21| : w2 e {- l,l}} , |z3 + = maxjz3 + w4z4| : w4 e {-1,1} IK + w6z6| = maxJK + w6z6| : w6 e {- 1,1} , IK + z &\\ = max K + w8zg|| : ws e {-1,1}
(4 bits needed). In the second stage sum vectors z^ t^and Z 5+^; 6 Z 6 are left untouched and we weight byw3 and w5 sum
vectors z3 + w4z4 and z7 + w8z8 in such a way that
||z, + w2z2 + w3(z3 + w4z4)|| = maxjz, + w2z2 + w3(z3 + w4z4)| : w3 e {1,7,-1,-7}}
+
Ή 6 +
zι + s
zΛ = maxj|z
5 + w
6z
6 + w
5 (z
7 + w
7z
7 )|| : w
5 e {1,7,-1,-7}}
( 4 bits heeded) . In the last stage sum vector z, + w2z, + w3(z3 + w4z4 ) =: w, is left untouched and we weight by wη the sum vector z5 + w6z6 + w5 (z7 + w&zs) -. u2 in such a way that
+ w7w2|| = max|K + w7w2|| : w7 = eM" )l n e {l,2,...,8}
( 3 bits needed) .
Assumptions for the third algorithm: M transmit antennas, the number of feedback bits is given in the description of the algorithm.
Third algorithm. In the first stage we divide all antennas into £, groups of Mk , k = 1,2, ... ,K] antennas . Here we can view each group of Mk vectors as a single (sum) vector and apply first or second algorithm. The number of feedback bits that are needed is
In the next stage we form new groups of Mk_ , k = 1,2, ... , K2 vectors in a manner that is known for both transmitter and receiver. Again we can view each group of Mk vectors as a single (sum) vector and apply first or second algorithm. If
we finish after L stages we find that the number of needed feedback bits is
The groups are preferably formed from stage to stage by spreading the members of each group (of previous stage) to as many different groups (of present stage) as possible.
Example (4 Tx antennas, 2 FB bits) . First and second vectors are left untouched in the first stage. Third and fourth vectors are weighted by w, in such a way that
||z, + z2 + w, (z3 + z4 )|| = maxjz, + z2 + w, (z3 + z4 )|| : w, e {- 1,1} }
( 1 bit needed) . In the second stage the sum vector z2 A- wiz3 is left untouched and the sum vector z Λ- w z4 is weighted by w2 in such a way that
|K + w,z3 + w2(z, + w1z4)| = max|K + wlz3 + w2 (z, + w,z4)| : w2 e {- l,l}
( 1 bit needed) .
Remark. If feedback consists (for example) of one bit at a time then the above algorithm (or similar algorithm that were given previously) can used continuously in time and only 2 channels need to be estimated simultaneously at any time.
All three main algorithms described above for phase rotations can be combined with each other.
In the following, three examples will be studied in more detail when z are independent identically distributed Gaussian random variables.
In the first example, a system with four transmit antennas is examined. The second algorithm is used and it is assumed that antennas are divided into groups of -two. Furthermore-,- we assume that M+2*N feedback bits are available: information from first N bits is used inside the first group, information from the following N bits is used inside the second group and information from the last M bits is used in the second stage when groups are rotated. The following SNR improvements have been obtained. Note that the number of needed feedback bits is in brackets after the SNR improvement.
Table 1.
The optimal SNR improvements - that can be achieved when only phases are adjusted and all possible combinations of different rotations determined by the possible FB words are checked - are listed in the following table.
It should be noted that the SNR improvement achieved by the algorith (s) in accordance with the invention is very near to optimal .
In the second example a system with 3 or 4 transmit antennas is examined. The first algorithm is used. Furthermore, it is assumed that (M-1)*N feedback bits are available: information from the first N bits is used when second phase is rotated, information from the following N bits is used when third phase is rotated and finally, information from the last N bits is used when fourth phase is rotated. The following table of SNR improvements has been obtained when different numbers (2*N or 3*N in total) of feedback bits are used.
The optimal SNR improvements - that can be achieved if all possible combinations of different rotations are checked - for the case M=3 are listed in the following table.
From the above results one sees that the above new algorithms give SNR improvements that are very near to optimal, and are furthermore extremely simple to apply.
In the third and final example we study the case of four transmit antennas. The third algorithm is used (see example corresponding to the case of three feedback bits) . At each of the three stages we have 1 or 2 feedback bits which are used in order to rotate the phase corresponding to the groups of 2 antennas. The following SNR improvements have been obtained. Note that the number of needed feedback bits is in brackets after the SNR improvement.
From the above table we see'that thi's method is not perfectly optimal. However, only two channels need to be estimated during each stage/time instant.
In the following , feedback algorithms corresponding to the power adjustment will be described.
Assume that the transmitted signal has the form
Z = ∑wkzk , zk = (zk(ϊ),zk(2),...,zk(L)\ zk(l) = k(l)eJ^l) =
where phases φk (l) and amplitudes ak (l) are known to the receiver. The aim is to adjust the weights wk >0 in such a way that
R = Zi
is maximized. It is assumed here that phase and power adjustments are done in the following manner: if there is P different power combinations available we adjust phases corresponding to each individual power combination and calculate the value of R for all p=l,2,...,P. From these P different values of R e then choose the one which gives the
maximum. The feedback bits corresponding to the best adjustment are signalled to the base station. In this method the phase adjustments can be done by applying any known scheme suitable for that purpose (including the algorithms of the previous section) . When there is two transmit antennas, three feedback bits for phase adjustment and a single feedback bit for power adjustment (as in WCDMA FDD mode 2), then the power adjustment alternatives are (0.2,0.8) and (0.8,0.2). If the resolution of the phase adjustment changes then also the power adjustment points should be changed. In our examples we have used power adjustment points that are valid when there are three feedback bits available for each phase adjustment. If this assumption is no longer valid, then new power adjustment points should be chosen. Next, some example algorithms for a four transmit antenna system are presented. First we set:
( 1 ) vc, + w2 + w3 + w4 ~ a (α, + br ) + b2 (c, + d )
where
and it is assumed that
a + b, = 1, c, + d} = 1, a2 + b2 = 1.
Now, values for the weight pairs (aλ ,bλ), (cλ,d ) can be chosen (for example from the set (0.2,0.8), (0.8,0.2) ) such that
(2 ; R, = max' β, z, + b,z2 (βl,A,) e {(0.2,0.8),(0.8,0.2)}
( 3 : R2 = max z3 + τjdl z4 (c ,d{ ) {(0.2,0.8), (0.8,0.2)}
It should ne noticed that in both of the above cases the maximum is taken over two different power adjustment alternatives after some phase adjustments has been done. For example, we can use algorithm I of the previous section for phase adjustments and then choose the best power adjustment by using the above two equations. From this operation we get two feedback bits for the power adjustment. In the next stage the maximization problem is solved:
(4) R = : (a
2,b
2) e {(0.3,0.7),(0.7,0
where we have first adjusted the phase between the sums ■ja2 (ψ2l zi + Λjb^z2 ) and ^b2 ψ:i zi + ^d\ Z4) corresponding to both power alternatives and then taken the maximum (now we get one feedback bit for power adjustment) . Note that the sets of different weight pairs need not be the same at each maximization problem. For example in problems (2), (3) we can have different sets than in the problem (4). In addition, the magnitude of sets -which corresponds to the number of feedback bits- can vary from stage to stage. Note that the above algorithm is closely related to the above mentioned second algorithm corresponding to the phase rotations.
In a similar manner we can design an algorithm that is based on the same structure as the above first algorithm corresponding to the phase rotations. Consider a formula
(5) w2 + w2 A- w2 + w2 = a (a2(aλ +bl ) + c, ) A- dl
where
w, -7 aa2a3 , w. =vb xa a3, w, S< w, =
and let us assume that
<2| +b, =l, <22 +e, =1, #3+<5?,=l,
Here we first find a pair (α,,b|) (for example from the set (0.2,0.8), (0.8,0.2) ) such that
R, = maxl /α,z,
: (α„δ
1)e{(0.2,0.8),(0.8,0.2)} is solved when phase adjustments has been done corresponding to both alternatives. From this operation we get one feedback bit for power adjustment. In the next stage, a maximization problem is solved
+ Jc
λ ~z
3 : (α
2,c,)e {(0.2,0.0.8),(0.8,0.2)}
where we have first adjusted the phase between
and sjc
lz
3 corresponding to both power alternatives and then taken the maximum (one feedback bit for power adjustment is obtained) . In the final stage we solve the problem
R = max ■ a3 ~Qaa2z + -Jb,a2z2 + Jc~z3 ) + Jd~z4 : (a3,dx)e {(0.2,0.8), (0.8,0.2)}
in the same manner as previous two problems
The third algorithm is closely related to the above discussed third algorithm for phase rotation. In the previous case of four transmit antenna example we solve first the equation:
R, = max α, (z, + z2 ) + Λ 3+z4 ).||: (α„δ,)e {(0.1,0.4), (0.4,0.1)}
where we have first adjusted the phase between the sums
and /b,(z
3+z
4) corresponding to both power alternatives and then taken the maximum (one feedback bit for power adjustment is given) . In the second stage, it is denoted z,
and we solve the equation:
R
2 + /b7(z,+z
4) : (a
2,b
2) e {(0.1,0.4), (0.4,0.1)}
where we have first adjusted the phase between the sums ^a
2(z
2 + z
3) and jb
2(z
l + z
4) corresponding to both power alternatives and then taken the maximum (one feedback bit for power adjustment is given) . In the third stage, it is denoted
'
Z4 jb
2z
4 and we solve the problem:
R
3 +z
3) + Jb
~ 3(z
2 +z
4) : (a
3,b
3) e {(0.1,0.4), (0.4,0.1)}
where we have first adjusted the phase between the sums ja
3(z
]+z
3) and
corresponding to both power alternatives and then taken the maximum (one feedback bit for power adjustment is given) . This algorithm can be continued in a periodic manner.
Both above-mentioned algorithms can be generalized in an obvious manner. In general we first decide the number of stages and the number of groups (and magnitudes of groups) at
each stage. This is due to the number of feedback bits. Then we form equation (as in (1) and (5) ) from which the weights can be computed. Feedback bits are obtained during computing the algorithm.
It should be noticed that in all examples this far it has been assumed that there are three feedback bits available for each phase adjustment. If this assumption is not valid then the power adjustment points must be changed.
Two or all three of the above main algorithms corresponding to the power adjustment can be combined with each other.
Further, one, two or all three of the above main algorithms corresponding to the power adjustment can be combined with one, two or all three of the above algorithms corresponding to the phase rotations.
In the following, some additional examples for applying the described new classes of feedback algorithms for transmit antenna diversity are given.
Consider the following transmit diversity system: 4 Tx antennas, a single receive antenna, Tx antennas 1, 2 form a pair in which symbols are transmitted with a delay (as in 2 Tx delay diversity) and Tx antennas 3, 4 form another pair in a similar manner. Thus symbols sl,s2,s3, ... are transmitted in the following manner
Here we can use feedback in order to adjust pairs of antennas against each other. Thus, in the receiver we estimate two channel parameters: one corresponding to pair (Txl,Tx2) (denoted by hi) and one corresponding to pair (Tx3, Tx4) (denoted by h2) . Then we apply the above described feedback algorithms in accordance with the present invention. The above-mentioned scheme can be extended directly to 2*M Tx antennas. We just divide antennas into M pairs, estimate M channels corresponding to the M pairs and apply previously given feedback algorithms.
As a further example, in the 4Tx example system given above we can also choose the better pair (in terms of received power) of two possible alternatives (1 FB bit is needed) and then transmit total power from the chosen pair. If 2*M pairs are used we need at least log2 (M) FB bits if the information about the best possible pair is signalled to the transmitter.
The invention is particularly suitable for 3GPP WCDMA FDD mode.
The invention describes a system-level concept preferably used within a wireless communication system. The base stations and mobile stations are adapted to interpret the the feedback messages in the same manner.
An optimization of feedback commands can be achieved, by taking into account a number of previously sent commands when more than 2 tx antennas are present - providing effectively a joint optimization of multiple feedback bits.
The performance of the currently proposed wcdma TX diversity feedback mode can be improved for M>2 tx antennas. Different ways are provided according to which the terminal can calculate the feedback.
Some simulations results for some cases in accordance with the invention showed high gains. Even with gains in the order of 0.2 dB - 0.5 dB it is possible, among others, to relax the terminal implementation margins elsewhere.
In general, [computationally] efficient improvements to the feedback modes are relevant, as they are directly related to implementation of the WCDMA terminal.