Sie sind auf Seite 1von 6

ISITA2010, Taichung, Taiwan, October 17-20, 2010

MIMO Systems with Mutual Coupling: How Many Antennas to Pack into Fixed-Length Arrays?
Shuo Shen, Matthew R. McKay and Ross D. Murch
Department of Electronic and Computer Engineering The Hong Kong University of Science and Technology Clear Water Bay, Kowloon, Hong Kong Email: eessx@ust.hk, eemckay@ust.hk, eermurch@ust.hk

AbstractFor multiple-input multiple-output wireless systems, we investigate the important question of how many antennas to place in xed-length arrays, accounting for both spatial correlation and mutual coupling at the transmitter and receiver. We show that there is an optimal antenna conguration yielding the highest capacity which depends strongly on the array length and the transmission wavelength, but does not depend strongly on the signal to noise ratio (SNR). Moreover, we show that ignoring the effect of mutual coupling gives misleading results, yielding unbounded capacity growth. As another key nding, we demonstrate the surprising result that if the optimal antenna conguration is employed, then further optimizing the transmission based on the channel statistics gives very little benet over simple equal-power spatial multiplexing. By deriving an expression for the capacity at low SNR, we provide a straightforward method for estimating the optimal number of antennas. Index TermsMIMO, correlation, mutual coupling, capacity

antenna patterns of the individual antennas themselves. In this paper we consider the following key issue which arises in the design of MIMO systems: How many antennas should be employed within the transmit and receive arrays of a xed length, in order to maximize capacity? This important practical question has yet to be adequately addressed. Some related work has been presented in [10] and [11], however the effects of mutual coupling were not explicitly accounted for, and a heuristic approach was employed to capture the inherent power loss. In this paper, we consider the rigorous mutual coupling model advocated in [7], and aim to gain practical insights into the channel capacity under various conditions. We rst present simulations to characterize how the MIMO capacity with spatial correlation and mutual coupling varies with the number of antennas for uniform linear arrays constrained to a xed length. Assuming that the transmitter has no knowledge of the channel and therefore employs equal-power spatial-multiplexing transmission, we nd that the capacity increases with up to a point opt , and then decreases beyond that point. In contrast, if mutual coupling is ignored, the capacity increases monotonically with and therefore gives misleading results, particularly beyond opt . We also investigate the effect of transmitter optimization, assuming that the channel statistics are known at the transmitter. Our key nding is that whilst transmitter optimization leads to noticeable capacity improvements beyond opt , for opt the benet is almost negligible, despite demanding a signicantly increased complexity. This demonstrates that for MIMO systems with xed length arrays, it is extremely important to optimize the number of antennas; however, if this is done correctly, then further optimization of the transmitted signals based on the channel statistics is not required. Whilst obtaining an analytical solution for opt as a function of the system parameters is difcult in general, we show that this problem can be simplied by considering the low signal to noise ratio (SNR) regime. To this end, we rst derive a simple expression for the low SNR capacity, adopting the general framework from [12, 13]. We then demonstrate that the antenna optimization problem can be re-posed into one involving the minimum required normalized energy per information bit, which admits a very simple expression and is trivial to evaluate.

I. I NTRODUCTION Multiple-input multiple-output (MIMO) technology is a key candidate for enhancing the capabilities of modern wireless communication systems. Two of the most important factors which govern the performance of practical MIMO systems are spatial correlation and antenna mutual coupling [1]. Spatial correlation occurs due to a lack of scattering in the environment, and its effect on MIMO performance has been studied extensively in recent years. Mutual coupling, on the other hand, is a physical phenomenon which occurs when the antennas becomes sufciently close to one another [2]. Compared with spatial correlation, the effect of mutual coupling has received far less attention. Prior work dealing with mutual coupling is presented in [36], where it was demonstrated that mutual coupling among closely-placed antenna elements leads to a lower correlation, and may in fact deliver increased capacity by providing a higher angle diversity. Further results were presented in [7] and [8], which investigated the effect of mutual coupling on the capacity of MIMO systems. In [7], the formulation is based on using antenna patterns in which the other ports are open, whereas in [8] the formulation is based on patterns in which the other ports are shorted. While both formulations are correct we prefer to use the formulation in [7]. This is because when we calculate the patterns with the other ports open the other antennas are effectively invisible [9] and the mutual coupling effects are fully incorporated into the coupling matrix itself. Therefore we can simply use the

9781424460175/10/$26.00 c 2010 IEEE 531

II. S YSTEM M ODEL Dipole antennas are assumed to be placed in an array of xed total length . The length of each dipole is a half wavelength. The same antenna arrays are used at both the transmitter and receiver. Letting and denote the number of transmit and receive antennas respectively, the 1 received signal vector is given by r = Hx + n, (1) where x is the 1 transmitted symbol vector, n is an 1 noise vector with independent zero-mean unit-variance complex Gaussian entries, and H is the channel matrix. We model H stochastically, taking into account the joint effects of spatial correlation and mutual coupling, as follows H = C R 2 H S 2 C .
1 1

The channel matrix can then be written as H= C R 2 H S 2 C ,


1 .
1 1

(7)

and the normalization factor =

III. MIMO C HANNEL C APACITY WITH M UTUAL C OUPLING AND C ORRELATION In this section, we investigate the capacity of the MIMO channel, taking into account both spatial correlation and mutual coupling effects. We also present analytical results for the low SNR regime, which lead to a simple method for optimizing the number of antennas to place in the transmit and receive arrays. A. Mutual Coupling and Optimal Number of Antennas In this section, we ask the question: If the transmitter and receiver have the same length- antenna array, and are each equipped with antennas, then what is the optimal number of antennas to maximize capacity? We start by assuming that H is known perfectly at the receiver (e.g., through the use of standard channel estimation techniques), but the transmitter either has no knowledge of H or chooses not to use this knowledge. In this case, it is common to assume that the input x is zero-mean complex Gaussian with covariance Q = I, which transmits independent equal-power signals from each antenna. In this case, the ergodic capacity1 (in bits/s/Hz) is given by [ ( )] SNR HH , iid = log2 det I + (8) where SNR is the average received SNR per element in the absence of mutual coupling. The expectation is taken with respect to the matrix-variate distribution of H. For our channel model (7), closed-form solutions can be obtained for this expectation based on the results in [15], with some minor modications. The equations however, involve considerable notation and are not reproduced here. Fig. 1 shows the capacity as a function of , for a total array length of 1 wavelength, i.e., = . The curve marked correlation and coupling is based on (8) and our channel model (7). The important observation from the gure is that if we explicitly take into account both the correlation effects of the channel and the mutual coupling effects of the antennas, then the capacity increases monotonically with up to a certain point, = 4, and then begins to decrease monotonically with . Fig. 2 shows a similar phenomenon for an array of length = 2; however, in this case the turning point is increased to = 6. The results shown in Figs. 1 and 2 are given for a moderately high SNR of 20 dB. Fig. 3 demonstrates the corresponding results for a low SNR of 0 dB, considering an array length of = . We observe the same general behavior as for the high SNR results and, perhaps surprisingly, we see the same turning point = 4 as shown in
1 Strictly speaking, this is the mutual information rather than the channel capacity, under the assumption of independent equal-power Gaussian inputs.

(2)

Here, H is an matrix of independent zero-mean unit-variance complex Gaussian variables, R is the correlation matrix at the receiver side, and S is the correlation matrix at the transmitter side. Assuming a uniformly distributed angular spread, from Jakes model, each entry of the receiver correlation matrix is given by R = 0 (2 /) , = 1, ..., , (3) where is the wavelength, 0 () is the zeroth-order Bessel function, and is the distance separating the th and th antenna elements. Assuming that the antenna elements are placed uniformly in the array, = 1 . The transmit correlation matrix S is obtained similarly. The matrix C and matrix C reect the mutual coupling effects at the receive and transmit sides respectively, and are dened according to [7] C = Z (Z + Z )1 , C = (Z + Z )1 . (4) Here, Z is an impedance matrix at the transmitter end with diagonal elements containing the self impedance terms and off-diagonal entries containing the mutual impedance terms, and Z is a diagonal matrix with diagonal entries equal to the conjugate of the diagonal entries of Z . For a dipole array, Z is well-known from standard antenna theory (e.g., see [14, Ch. 8]). Similar to Z , Z is the impedance matrix at the receiver end, and Z is a diagonal matrix with diagonal entries equal to the conjugate of the diagonal entries of Z . The factor in (7) is used for normalization purposes. To establish this constant, we note that if the receiving antenna elements are uncoupled, then C is diagonal with entries [C ] = (Z ) (Z )11 = = . (Z ) + (Z ) (Z )11 + (Z ) 11 (5)

Hence, it makes sense to normalize the receiver mutual coupling matrix C by dividing by . Similarly, we introduce a normalization factor for the transmitter, 1 . (6) = (Z )11 + (Z ) 11
532

Fig. 1. This implies that optimal number of antennas to employ does not have a strong dependence on the operating SNR, but as expected has a strong dependence on the relative antenna spacing. This, in turn, suggests that we may gain more insights into the optimal antenna conguration by focusing on these asymptotic regimes, in which case the capacity expression (8) simplies into more intuitive forms. This will be explored in more detail in Section IV, where we focus primarily on the low SNR regime. B. Comparison with Correlation-Only Model Whilst much previous work has investigated the effect of spatial correlation on the capacity of MIMO systems (see e.g., [15, 16]), the effect of mutual coupling is typically not considered. Thus, it is of interest to study the difference in capacity between such correlation-only models, and the capacity with mutual coupling. The correlation-only model is obtained from (7) by substituting C = I and C = I, in which case the channel then reduces to the now-standard representation 1 1 H = R 2 H S 2 . The corresponding curves are shown in Figs. 13, labeled as correlation but no coupling. From these curves, we observe some very interesting behavior. First, for the high SNR scenarios in Figs. 1 and 2, the correlation-only curve closely follows the corresponding correlation-coupling curve up to the turning point, indicating that mutual coupling has relatively little impact if the SNR is high and the number of antennas is small. However, beyond the turning point, the behavior is markedly different. In particular, the correlationonly curve, whilst having a deection at the turning point, still continues to increase monotonically as the number of antennas are further increased. Thus, beyond the turning point, considering a correlation-only gives incorrect results, implying that the capacity grows unbounded with . Similar trends are also seen for the low SNR scenario in Fig. 3, however in this case the correlation-only model overestimates the capacity even for below the turning point. The difference in behavior observed by ignoring mutual coupling can be explained by comparing the average channel gains [tr(HH )] for spatially correlated channels, with and without mutual coupling. We know that mutual coupling will yield a loss in transmission power, which in our model is reected by a loss in the average channel gain. This is shown in Fig. 4, where the average channel gain with mutual coupling is seen to increase with up to the turning point ( = 4), and then decreases monotonically. If mutual coupling is neglected however, the average channel gain grows unbounded. These results, in general, indicate the importance of considering both channel effects (spatial correlation) and antenna effects (mutual coupling) to accurately predict the capacity of MIMO channels. C. Benet of Transmitter Optimization So far, we have assumed that the transmitter employs uncorrelated equal-power Gaussian inputs. If the transmitter has knowledge of the channel distribution however, then it
533

30 correlation and coupling correlation but no coupling correlation and coupling with power allocation

25

Capacity (bits/s/Hz)

20

15

10

5 No. of antennas

10

15

Fig. 1. Comparison of ergodic MIMO Capacity with (i) mutual coupling and equal power allocation, (ii) no mutual coupling and equal power allocation, and (iii) mutual coupling and optimal power allocation. Results are shown for SNR = 20 dB and = 1.
40

35

30 Capacity (bits/s/Hz)

25

20

15 correlation and coupling correlation but no coupling correlation and coupling with power allocation 0 5 No. of antennas 10 15

10

Fig. 2. Comparison of ergodic MIMO capacity with (i) mutual coupling and equal power allocation, (ii) no mutual coupling and equal power allocation, (iii) mutual coupling and optimal power allocation. Results are shown for SNR = 20 dB and = 2.

can use this knowledge to optimize its transmission. In this case, it is well-known that the ergodic capacity is given by [ ( )] SNR opt = max log2 det I + HQH , (9) Q where the optimization is over the input covariance matrix Q = 1 [xx ] 2 ] , which is normalized to satisfy tr[Q] = . [x A closed-form solution to this problem is intractable, however iterative solutions have been proposed [17]. In general, the ergodic capacity (9) will be higher than that achieved with equal-power inputs in (8), however the complexity is also increased signicantly. Moreover, this complexity difference increases with the number of antennas. Here, we aim to investigate the capacity benet obtained through transmitter optimization, taking into account both mutual coupling and spatial correlation. The ergodic capacity curves based on (9) are plotted on Figs. 13, labeled as correlation and coupling with power allocation. These results were generated using the iterative algorithm from [17]. On all three gures we see the interesting behavior that, up to the respective turning points, the optimization gives very little benet. This is true for both high and low SNR. Moreover, even with transmitter optimization,

7 correlation and coupling correlation but no coupling correlation and coupling with power allocation

100 90 80 no mutual coupling mutual coupling

5 Capacity (bits/s/Hz) Channel Gain 0 5 No. of antennas 10 15

70 60 50 40 30 20

10 0 1 2 3 4 5 6 No. of antennas 7 8 9 10

Fig. 3. Comparison of ergodic MIMO capacity with (i) mutual coupling and equal power allocation, (ii) no mutual coupling and equal power allocation, and (iii) mutual coupling and optimal power allocation. Results are shown for SNR = 0 dB and = 1.

Fig. 4. Comparison of channel gain [tr(HH )] for (i) correlation and mutual coupling, (ii) correlation but no mutual coupling. Results shown for = 1.

IV. A NALYTICAL C HARACTERIZATION AT L OW-SNR we observe the same exact turning points as for the equalpower case, beyond which the capacity begins to decrease monotonically. Thus, the main benet of optimization is only seen to be a more gradual degradation with . This can be explained by noting that as increases, mutual coupling and correlation become more severe, and the transmitter can exploit this information to more effectively steer or beamform the signals along stronger channel directions. The expense paid for this more gradual degradation in capacity is a much higher complexity, which increases signicantly with . These results are quite unexpected and highlight the importance of selecting the optimal number of antennas. In particular, they imply that if the system is optimized with respect to the number of antennas, then there may be very little to be gained from covariance optimization of the input signals. It is therefore of interest to further investigate analytical methods for establishing this optimal antenna conguration, for a given antenna array of a given length. In Section IV we will do this by appealing to the low SNR regime. D. Comparison with Previous Models Previous contributions have shown that for a MIMO system with xed length antenna arrays and only correlation, the capacity converges to a limit as goes to innity [10, 11]. In their work, the following approximation was used to model the effect of xed length arrays: )] [ ( SNR = log2 det I + HH . (10) 2 This modication, which further scales the transmitted power by 1/ , ensures that the total received power does not diverge as increases, and provides a straightforward approximation for capturing the power loss effect caused by mutual coupling. However, we have found that whist this model yields the same turning point as the complete correlation-coupling model in (7), beyond this turning point the approximation is no longer accurate. We also mention that the alternative mutual coupling model presented in [8] (based on shorted ports), seems to yield a different turning point altogether.
534

From the results above, we have observed that there is a capacity turning point as the number of antennas increase, and that this turning point depends primarily on the total length of the antenna array (relative to the wavelength), but not the SNR. Thus, to gain more insights into the joint effect of correlation and mutual coupling, here we focus on the low SNR regime. A. Low-SNR Capacity with Mutual Coupling and Correlation A general framework for studying the low SNR capacity was presented in Verd s pioneering paper [12], and further u elaborated for MIMO systems in [13]. It was shown that at low SNR the capacity could be expressed as a linear function of the normalized energy per information bit, , as follows 0 ( ) ( ) (11) C 0 log2 0 0
0

min

where denotes the minimum required for reliable 0 communications, and 0 is referred to as the wideband slope. Based on this, we have the following key result: Proposition 1: For a MIMO system with mutual coupling and correlation, the capacity at low SNR, as a function of , 0 is given by (11) with

0 min

2 2 = log 2 0 min tr( )tr( ) and 0 = 2


tr(2 ) tr2 ( )

(12)

tr(2 ) tr2 ( )

(13)

where = C RC and = C SC . Proof: See the Appendix. Fig. 5 compares the exact capacity, based on Monte-Carlo simulation, with the linear approximation based on Proposition 1. As evident from the gure, the analytical approximation is quite accurate. Note that the results are shown as a function of 0 , where is the received energy per information bit in the absence of mutual coupling, which is given by = , with dened in (19).

B. Comparison with Correlation-Only Model We now investigate the implications of ignoring mutual coupling at low SNR. We require the following corollary: Corollary 1: For a MIMO system with correlation but no mutual coupling, the capacity at low SNR, as a function of 0 , is given by (11) with 1 = log 2 0 min and 0 = 2
(S)

8 7 6 Capacity (bits/s/Hz) 5 4 3 2 N=1 exact N=1 appx N=4 exact N=4 appx

(14)

1 0 2

0 Eb/N0 (dB)
r

(R)

(15)

Here, () is the matrix dispersion function, dened for an matrix X as (X) = tr(X2 )/. Proof: Follows by setting C = I and C = I in Proposition 1, and using tr(R) = and tr(S) = . The result in Corollary 1 was derived previously in [13]. Comparing (12) and (14), we see that whilst correlation by itself has no effect on min , if both mutual coupling and 0 correlation are considered, then min has a strong depen0 dence through the matrices and . In particular, mutual coupling has the negative effect of increasing the required 0 min by a factor of , as given in (21). This behavior is illustrated in Fig. 6, which plots min as a function of the 0 number of antennas, considering = = . The curves are based on (12) and (14). We clearly see that if mutual coupling is ignored, then min decreases monotonically as 0 more antennas are added. With mutual coupling, however, the behavior is markedly differentdecreasing to a point = 4, and then increasing beyond that. This is explained intuitively by noting that for 4, the additional multiplexing gain afforded by adding more antennas outweighs the power loss due to mutual coupling, as the number of antennas increase. Beyond this, however, the power loss due to mutual coupling has a more dominant effect (i.e., due to the antennas being packed so closely together), which leads to a net loss in overall capacity. Importantly, this turning point is consistent with that found for the exact capacity in Figs. 1 and 3. These results again highlight the importance of considering a joint correlationcoupling model. Fig. 7 plots the wideband slope 0 versus the number of antennas, comparing the scenarios with and without mutual coupling. Again, = = is considered, and the curves are based on (13) and (15). We see that in contrast to the min , mutual coupling has relatively little effect on 0 0 . Perhaps surprisingly, the wideband slope is underestimated if mutual coupling is ignored. The most important point, however, is that we see the same general trend with and without mutual coupling, and that the familiar turning point = 4 occurs for both curves. C. Estimating the Optimal Number of Antennas Whilst it is very difcult to nd the optimal number of antennas to maximize the capacity, even at low SNR, based on
535

Fig. 5. Comparison of exact ergodic MIMO capacity with mutual coupling and correlation, and the analytical low SNR approximation. Results are shown for = 1 and = 4, with = 1.
0

4 Eb/N0 min(dB)

10

correlation and coupling correlation but no coupling 1 2 3 4 5 6 No. of antennas 7 8 9 10

12

Fig. 6.

Comparison of the minimum required energy per information bit for (i) correlation and mutual coupling, and (ii) correlation but no mutual coupling. Results are shown for = 1.
0 min

Fig. 7 (and also on other numerical experiments not shown), we propose a new simple design method for estimating the optimal solution. The method is to consider a correlationonly model, and to choose the number of antennas which maximizes 0 . For the case of = = , this leads to the very simple optimization problem: tr(R2 ) . (16) 2 Whilst a closed-form solution is not forthcoming, this is trivially evaluated numerically for any given value of and , based on (3). The results obtained through this optimization have been validated experimentally, and have been found to accurately estimate the optimal capacity-maximizing antenna conguration. opt = arg min V. C ONCLUDING R EMARKS We have investigated the capacity of MIMO wireless systems where dipole antennas are placed in a xed-length linear array. We have demonstrated that it is critical to account for the joint effects of spatial correlation and mutual coupling, and that ignoring the effects of mutual coupling gives misleading results. Another important and unexpected nding is that if the number of antennas is properly chosen, then further optimizing the transmitted signals based on the channel statistics yields

and 0 = 2[C(0)]2 log 2 . [C(0)] (23)

3.5

3 Slope

2.5

d Using the properties d log det [I + A]=0 = tr(A) and d2 2 d2 log det [I + A]=0 = tr(A ), we have
correlation and coupling correlation but no coupling

1.5

log 2 = 0 min [tr(H1 H )] 1


10

(24)

5 6 No. of antennas

and 0 = (H1 H ) 1 2 . (25)

Fig. 7. Comparison of the wideband slope 0 in bits/s/Hz/(3dB) for (i) correlation and mutual coupling, and (ii) correlation but no mutual coupling. Results are shown for = 1.

From (21) and (24) we obtain (12). From (18) and (25) we obtain (13). R EFERENCES
[1] Z. Xu, S. Sfar, and R. S. Blum, Receive antenna selection for closelyspaced antennas with mutual coupling, IEEE Trans. Wireless Commun., vol. 9, no. 2 pp. 652-661, Feb. 2010. [2] J. L. Allen and B. L. Diamond, Mutual coupling in array antennas, Lincoln Laboratory, M.I.T., Lexington, MA, Tech. Rep. 424 (ESD-TR66-443), 1966. [3] J. Luo, J. R. Zeidler, and M. A. Jensen, Performance analysis of compact antenna arrays with MRC in correlated Nakagami fading channels, IEEE Trans. Veh. Technol., vol. 50, no. 1, pp. 267-277, Jan. 2001. [4] M. Stoytchev, H. Safar, A. L. Moustakas, and S. Simon, Compact antenna arrays for MIMO applications, in Proc. IEEE Antennas and Propagation Society Int. Symp., vol. 3, Boston, MA, July 8-13, 2001, pp. 708-711. [5] C. Waldschmidt, J. v. Hagen, andW. Wiesbeck, Inuence and modeling of mutual coupling in MIMO and diversity systems, in Proc. IEEE Antennas and Propagation Society Int. Symp., vol. 3, San Antonio, TX, June 16-21, 2002, pp. 190-193. [6] J. W. Wallace and M. A. Jensen, Termination-dependent diversity performance of coupled antennas: Network theory analysis, IEEE Trans. Antennas Propagat., vol. 52, no. 1, pp. 98-105, Jan. 2004. [7] J. W. Wallace and M. A. Jensen, Mutual coupling in MIMO wireless systems: a rigorous network theory analysis, IEEE Trans. Wireless Commun., vol. 3, no. 4, pp. 1317-1325, July 2004. [8] R. Janaswamy, Effect of element mutual coupling on the capacity of xed length linear arrays, IEEE Antennas Wireless Propagat. Lett., vol. 1, no. 1, pp. 157-160, 2002. [9] R. G. Vaughan and J. B. Andersen, Antenna diversity in mobile communications, IEEE Trans. Veh. Technol., vol. VT-36, pp. 147-172, Nov. 1987. [10] S. Wei, D. Goeckel, and R. Janaswamy, On the asymptotic capacity of MIMO systems with antenna arrays of xed length, IEEE Trans. Wireless Commun., vol. 4, no. 4, pp. 1608-1621, July 2005. [11] R. Couillet, S. Wagner, M. Debbah, and A. Silva, The space frontier: Physical limits of multiple antenna information transfer, in Proc. ICST Conference on Performance Evaluation Methodologies and Tools, Athens, Greece, Oct. 2008. [12] S. Verd , Spectral efciency in the wideband regime, IEEE Trans. u Inform. Theory, vol. 48,no. 6, pp. 1319-1343, June. 2002. [13] A. Lozano, A. M. Tulino, and S. Verd , Multiple-antenna capacity in u the low-power regime, IEEE Trans. Inform. Theory, vol. 49, no. 10, pp. 2527-2544, Oct. 2003. [14] C. A. Balanis, Antenna Theory: Analysis and Design. New York: Wiley, 1997. [15] H. Shin, M. Z. Win, J. H. Lee, and M. Chiani, On the capacity of doubly correlated MIMO channels, IEEE Trans. Wireless Commun., vol. 5, no. 8, pp. 2253-2265, Aug. 2006. [16] M. R. McKay and I. B. Collings, General capacity bounds for spatially correlated Rician MIMO channels, IEEE Trans. Inform. Theory, vol. 51, no. 9, pp. 3121-3145, Sep. 2005. [17] A. M. Tulino, A. Lozano, and S. Verd , Capacity-achieving input u covariance for single-user multi-antenna channels, IEEE Trans. Wireless Commun., vol. 5, no. 3, pp. 662-671, Oct. 2003.

very little additional gainsimply employing equal-power spatial multiplexing techniques yield almost the same capacity. We also identied a simple method for estimating the optimal antenna conguration for a given array length and transmission wavelength, based on capacity expressions which we derived for the low SNR regime. A PPENDIX We start by noting that, due to the effect of mutual coupling, there is a power loss at the transmitter and the receiver. Thus, whilst the channel gain [tr(HH )] without mutual coupling is , with mutual coupling it becomes less. It is therefore convenient to dene a normalized matrix H1 , satisfying [tr(H1 H )] = , 1 (17)

in order to separate the distinct effects of power loss (due to coupling) and correlation. From (7), this can be written as H1 = 0 C R 2 H S 2 C ,
1 1

(18)

where 0 is an appropriate normalization factor, and the channel matrix H becomes H = H1 . Dene as the 0 average power loss, which is given by = From (17) and (18),
2 0 =

[tr(HH )] [tr(H1 H )] 1

2 . 0

(19)

[tr(C RC H C SC H )] . = tr( )tr( )


1

(20)

Substituting =

and (20) into (19) yields tr( )tr( ) . 2 2 (21)

Now, from [12, 13], min and 0 can be computed from the 0 rst and second derivatives of (SNR) at SNR = 0 via

1 = 0 min C(0) [tr(H1 H1 )]

(22)

536

Das könnte Ihnen auch gefallen