Several mechanisms of radio resource management have effects on the system capacity and cell coverage of CDMA networks. In this section, we discuss the effects of pilot and soft handoff power allocation schemes. Some preliminary simulation results are also presented.
4.2.1 Effects of Pilot Power Allocation Schemes
Since each base station has finite power resource, the pilot channel and traffic channels have to share the total power resource. This explains the interdependence of coverage and capacity in CDMA systems. Pilot power allocation can be either fixed or dynamic. In fixed pilot power allocation scheme, which is used by current CDMA systems, about 10-15% of the total power is allocated to the common pilot channel and is not changed after the deployment of a cellular network, as shown in Fig. 4.1(a). Fig. 4.1(b) shows the dynamic pilot power allocation scheme. With the maximum and minimum constraints, the pilot power can be
h Rµ RM
M µ
LM,h Lu,h
SSDT
(c) Power allocation during soft handoff pM,h=0.0
pu,h=3.5
LPPA pu,h=2.8 pM,h=1.4 (b) Power allocation before soft handoff (a) Soft handoff in a 2-cell model
required power level
LM,h Lu,h pM,h=7.0
pu,h=0.0
max. link power max. link power
Figure 4.2: Diagram of the soft handoff power allocation in downlink CDMA systems with (a) soft handoff in two mixed-size cell model, (b) before soft handoff, and (c) during soft handoff.
adjusted between them based on various traffic situations. When the required traffic is low, the pilot power can be increased to extend cell coverage so as to accommodate more users around the adjacent cells. On the other hand, when the required traffic power is too high to have risks of degrading system performance, the pilot power can be decreased to shrink cell coverage.
4.2.2 Effects of Soft Handoff Power Allocation Schemes
In this subsection, we illustrate the effects of soft handoff power allocation on system capacity and cell coverage by a simple example. Consider a handoff user h located around the boundaries of cell M and cell µ, which have different sizes (RM 6= Rµ), as shown in Fig.
4.2(a). The power allocated to the user before soft handoff is shown in Fig. 4.2(b). Fig.
4.2(c) shows the allocated power during soft handoff in the SSDT and LPPA schemes. The height of each block is the constraint of maximum link power for each cell, and the width is
the link quality, LM,h (Lµ,h), from cell M (µ) to the mobile station h. Assume Lµ,h = 2LM,h in the example. Moreover, let pM,h(pµ,h) denote the allocated link power from cell M (µ).
By multiplying the allocated link power with the link quality, we can obtain the received signal quality of the user, which is the area of the shadowed blocks in Fig. 2. Assume that the received signal quality is 14 (7 × 2) before soft handoff. During soft handoff, the SSDT scheme, defined in 3GPP [14], [46], allocates power level 3.5 units to the user to guarantee the same requested signal quality (3.5 × 4 = 14). Because power level 3.5 units exceeds the constraint of maximum link power 3.0 units, the user cannot be served. Consequently, the user is dropped and the coverage of cell µ is shrunk. In the LPPA scheme of Chapter 2, 3 and [8], transmission power is allocated to all links in the active set distributively, in which the power level is proportional to the link quality. In Fig. 4.2(c), pM,h= 1.4 units and pµ,h = 2.8 units. The received signal quality in LPPA is 14 (1.4 × 2 + 2.8 × 4). Therefore, the allocated power in each link does not exceeds the constraint of maximum link power. LPPA can still obtain the same requested signal quality. Thus, LPPA achieves larger service coverage and capacity than SSDT. This example can explain the possible impacts of soft handoff power allocation schemes on cell coverage and capacity.
4.2.3 Simulation Examples of Adjusting Pilot Power Only
The above discussions reveal that other radio resource management algorithms besides pilot power allocation have great impacts on dynamic cell configuration. In this subsection, we show by simulation examples that pilot power allocation and other radio resource man-agement algorithms are highly coupled. Dynamically adjusting pilot power alone while not changing other radio resource management algorithms accordingly can result in performance degradation. Detailed simulation environment and parameters will be given in Section 4.5.
Fig. 4.3 shows the system throughput with different pilot power levels are considered under uniform (ρ = 1) and (b) non-uniform (ρ = 4) cell load cases for both SSDT and LPPA soft handoff schemes. For the non-uniform cell load cases, the traffic load in the hotspot cell is 4 times that of the surrounding cells. Note that only pilot power is adjusted, and all other radio resource management algorithms are not changed accordingly in the simulations. In
1 1.5 2 2.5 3 3.5 4
Figure 4.3: Capacity results of the fixed pilot power design by applying SSDT and LPPA schemes under cases with (a) uniform (ρ = 1) and (b) non-uniform (ρ = 4) cell loads.
the simulations, 1 watt power is allocated to the pilot channel in the fixed pilot power allo-cation scheme, and all other radio resource management algorithms are optimized according to this pilot power setting and pre-defined link budget design. We further adjust the pilot power level manually but not changing all other radio resource management algorithms.
We can see from Fig. 4.3 that the system throughput degrades whenever a pilot power other than 1 watt (used in the fixed scheme) is used in dynamic pilot power allocation schemes. This result is not surprising, because pilot power can also cause interference to users. We can observe that the larger the pilot power, the larger the interference, and the lower the throughput. Another reason is that the call admission control criterion and the constraint of maximum link power remain the same when the pilot power changes cell coverage. For example, when new or handoff calls issue requests to cells with light (heavy) traffic load, the tight (loose) criteria of CAC may result in new call blockings or handoff forced terminations. The uncoordinated design of pilot power and other radio resource management strategies can degrade the system performance severely. It is also observed in Fig. 4.3 that soft handoff power allocation and pilot power allocation are highly coupled, both of which have effects on the system throughput. LPPA has larger throughput than
SSDT with different pilot power and traffic load distributions. The throughput difference between SSDT and LPPA is larger when the traffic load is non-uniformly distributed (ρ = 4).
4.2.4 Our Approach
This chapter proposes a novel dynamic cell configuration in next-generation CDMA net-works via reinforcement-learning technique, which takes into account pilot, soft handoff power and maximum link power allocations as well as call admission control mechanisms [56]. Fig. 4.4 shows the system block diagram of our proposed approach. Each base station is equipped with our proposed scheme. A base station adjusts its pilot power periodically to adapt to the variations of system situations through the dynamic pilot power controller.
Based on the determined pilot power level, the maximum link power constraint and call ad-mission control criterion are adjusted accordingly. Then, the traffic channel power allocator adjusts its constraint of maximum link power that is obtained from the maximum link power estimator. After applying all updates for radio resource management, the reinforcement sig-nal is input to the dynamic pilot power controller to aid its decision for the next pilot power level. The detailed design will be illustrated in the following sections.
4.3 System Model
In this section, we describe the signal model and the link budget model in CDMA systems.
Then, we introduce the problem of dynamic cell configuration.
4.3.1 Signal Model
Assume the total allocated power of a base station b is Pb, including pilot channel power PbI and traffic channel power PbT, where PbT is smaller than the base station’s maximum transmission power ePb. Furthermore, for the pilot power of base station b, PbI = fb × ePb, where fb is the fraction of the pilot power to base station b’s maximum transmission power, in which fb ∈ [fmin, fmax]; fmin and fmax are the minimum and maximum constraints of the pilot fraction. On the other hand, for the traffic channel power of base station b, assume φm is the allocated power ratio of the traffic channel power for mobile station m, so the allocated
Base Station
Figure 4.4: System block diagram of the proposed dynamic cell configuration (DCC) scheme with radio resource management.
power pb,m for mobile station m from base station b is pb,m = φmPbT ≤ epb, where epb is the maximum link power constraint. Thus, P
m∈Ubφm = 1, where the Ub represents the set of all mobile stations served by base station b.
Define Lb,m as the link quality between base station b and mobile station m, and ηo is the thermal noise. Note that the link quality depends on effects of both path loss and shadowing.
The total received interference of the mobile station m served by the base station b, Ib,m, is
Ib,m = (1 − fα)PbTLb,m+X
k6=b
PkTLk,m+ ηo, (4.1)
where fαis the orthogonality factor. Note that the first and second terms in (4.1) mean intra-cell and inter-intra-cell interferences, respectively, in which the first term is caused by imperfect orthogonality of channel codes.
The received chip-energy-to-interference ratio (Ec/Io), denoted by υb,m, for mobile station
m with service rate r served by base station b is
υb,m = PbILb,mGIP Ib,m
≥ Υ, (4.2)
where GIP is the processing gain of pilot signal, Υ is the minimum Ec/Io constraint of the system. In order to maintain good connection with at least one base station in the system, the mobile station’s Ec/Io should exceed Υ at all the time. In subsection 4.3.3, detailed designs of pilot power PbI and maximum link power constraint for service rate r, epb(r), in CDMA systems are discussed using link budget analysis.
Moreover, in CDMA cellular systems, for mobile station m with service rate r served by base station b, the received bit-energy-to-interference ratio (Eb/Io), denoted as γb,m(r), must be larger than or equal to the required service quality, denoted as γ∗(r). For bandwidth W , γb,m(r) of mobile station m can be expressed as
γb,m(r) = φmPbTLb,mGP(r)
Ib,m ≥ γ∗(r), (4.3)
where GP(r) = W/r is the processing gain for service rate r. Furthermore, for a soft handoff user h with service rate r, the maximum ratio combining (MRC) method is used to combine signals from all serving base stations in its active set Dh. Thus, its received Eb/No, γh(r), can be obtained by
γh(r) = X
b∈Dh
γb,h(r). (4.4)
4.3.2 Handoff Power Allocation Schemes
Soft handoff is one of the important features in CDMA cellular mobile communication systems. When mobile users move from one cell to another cell, the soft handoff technique can provide seamless connections and better signal qualities for users in the cell boundary.
Because of multiple links transmission, power balance can be achieved between cells by executing soft handoffs. As a matter of fact, the service coverage can thus be extended. In the following, two techniques of handoff power allocation are discussed.
A. The Site Selection Diversity Transmission (SSDT) Scheme
The main concept of site selective transmit diversity (SSDT) scheme [14] is to dynamically choose one base station with the best link quality in the active set for transmission in order to mitigate interference caused by multiple site transmission. Assume there is one handoff mobile station h with service rate r. Let pb,h(r), where b ∈ Dh = {1, · · · , |Dh|}, be the required transmission power from base station b for satisfying the required service quality of handoff user h with service rate r such that γh(r) ≥ γ∗(r), which can be calculated from (4.3). According to the SSDT scheme, the transmission power is allocated as follows
• [STEP1]: Select the best link κs among the links in the active set so that the base station can allocate the least transmission power:
κs= argb min©
p1,h, p2,h, · · · , p|Dh|,h
ª. (4.5)
• [STEP2]: The required transmission power, p∗h(r), for mobile station h from serving base station b is
p∗h(r) =
½ pb,h, if b = κs
0, if b 6= κs (4.6)
• [STEP3]: The allocated power from base station b should be confined by the constraint of maximum transmission power for each link
pb,h(r) = min { p∗h(r), epb} . (4.7)
It is noteworthy that because of the maximum link power constraint, the SSDT scheme sometimes cannot offer enough required power for soft handoff users.
B. The Link Proportional Power Allocation (LPPA) Scheme
The link proportional power allocation (LPPA) scheme was suggested in Chapter 2, 3 and [8].
The main idea of the LPPA scheme is based on the link proportional strategy to distribute required transmission power of the handoff user among all the links in the active set. That is, the base station with better (weak) link connection provides more (less) power. In other
words, the LPPA finds a set of pb,h, where b ∈ Dh = {1, · · · , |Dh|}, such thatP
b∈Dhγb,h(r) ≥ γ∗(r) and
p1,h : p2,h : · · · : p|Dh|,h = L1,h : L2,h : · · · : L|Dh|,h , (4.8) where Lb,h is the link quality from base station b to handoff user h. According to the LPPA scheme, the transmission power is allocated as follows
• [STEP1]: Calculate weighting factor of the required power from base station b, b ∈ Dh, for user h as
wb,h = Lb,h P
b∈Dh
Lb,h
. (4.9)
• [STEP2]: Calculate the required transmission power, pb,h(r), from base station b to the soft handoff user h by
pb,h(r) = min{ wb,h· p∗h(r), epb}, ∀ b ∈ Dh, (4.10) where p∗h(r) is the required total power that can be obtained using iterative method by the tuning ratio of required service quality to the resultant service quality, γ∗(r)/γh(r).
The LPPA scheme estimates the required power for soft handoff user h, p∗h(r); then it distributes p∗h(r) to all serving base stations in Dhunder the maximum link power constraint in base station b ∈ Dh, epb. If the required power of one link over the constraint of maximum transmission power, the LPPA compensates the required power through other links. The detailed design and the prove of convergent can be found in [8].
4.3.3 Link Budget Analysis
From the perspective of the base station’s transmitter, there exists antenna gain GB and cable loss LC of the base station, the equivalent isotropic radiated power (EIRP), EP, of the traffic channel can be calculated by:
EP[dBm] = epb[dBm] + GB[dBi] − LC[dB]. (4.11) Note that units of parameters are denoted by the bracket in the following descriptions 1. On the other hand, from the perspective of mobile station’s receiver, taking soft gain GS,
1In this chapter, if the unit of a variable is not specified, the variables is linear.
antenna gain GM, and body loss LD of the mobile station into account, the total EIRP, ET, is
ET[dBm] = EP[dBm] + GM[dB] − LD[dB] + GS[dB]. (4.12) Moreover, consider the budget of interference margin ΩI and received noise power ηo, the receiver sensitivities of the mobile station with different service rates is
HR(r)[dB] = HS(r)[dB] + (ΩI + ηo)[dBm], (4.13) where HS(r)[dB] is the required signal to interference and noise value for different service rate r, which is equal to required bit-energy-to-noise ratio, γ∗(r)[dB], minus processing gain G(r)[dB]. From the preceding link budget, leaving a margin for log-normal fading ΩL, the resultant allowable maximum pathloss for different service rates is
P L(r)[dB] = ET[dBm] − HR(r)[dB] − ΩL[dB]. (4.14) Based on the allowable maximum pathloss and the applied channel model, the resultant cell radius R(r) is different to service rates r. For r ∈ [ rmin, rmax ], from (4.13) and (4.14), since HS(rmin) > HS(rmax), then P L(rmin) < P L(rmax). We can thus find that R(rmin) > R(rmax) as shown in Fig. 4.5 such that different service rates have different service coverage. Apparently, this phenomenon arises the issue of fairness for different service rates since it’s unfair for the mobile station with higher service rates to be served near cell boundary. In order to take fairness of service coverage into account for the mobile station with different service rates near the cell boundary, it is necessary to choose a suitable reference service rate for the cell radius design. Assume the default cell radius is reference to service rate r∗, where r∗ ∈ [rmin, rmax].
In general, pilot power PbI is around 1 watt to 4 watt, which is about 5% − 20% of the maximum transmission power of base station b, ePb. When a mobile station at the cell boundary, the received chip-energy-to-interference ratio, Ec/Io, should be equal to or larger than the required Ec/Io, Υ, which can be calculated as
Υ[dB] = PbI[dBm] − P L(r∗)[dB] − (ΩI+ ηo)[dBm]. (4.15) Normally, Υ is within the range from −16 [dB] to −20 [dB].
R( rmin )
R( rmax )
Base Station Base Station
Figure 4.5: Service coverage with different service rates.
4.3.4 DCC Problems
In future WCDMA networks, diverse multimedia traffics and random user mobility make cell coverage and capacity difficult to design. This is because each base station has finite power resource to be shared among users, the allocated pilot and traffic channel power are directly related to the coverage and capacity of the cell. Conventionally, fixed pilot power scheme is adopted. However, when the traffic load is too high to allocate enough required power for users, the system performance will be degraded severely. Some strategies of radio resource management have to be employed to release power or balance power among cells. The dynamic pilot power scheme is one possible approach to balance power load among cells. Therefore, in order to achieve load balance whenever traffic congestion occurs, dynamic cell configuration by adjusting pilot power is necessary. When the required traffic power is low (high), the base station could increase (decrease) its pilot power level to extend (shrink) cell coverage so as to accommodate (release) more users around the adjacent cells.
However, dynamic cell configuration may induce some side effects on the system performance as described in the followings.
• For new arriving mobile stations near cell boundaries, their initial access cells will be determined by the received signal strength, which directly relate to the pilot power level of the base stations. Therefore, reducing the pilot power of the congested cell causes them to request traffic channels from other adjacent cells. If an initial new call fails to detect any base station with enough Ec/Io, it cannot make a call request to
the system. This is the so-called coverage failure. As a consequence, although the new call blocking probability of the congested cell could be decreased, the coverage failure probability might be increased.
• For existing mobile stations near cell boundaries, decreasing (increasing) the pilot power of a base station may force some of them to handoff into other (its) cell(s).
Therefore, the average size of the active set and handoff rates would be increased. In addition, if a mobile station suffers bad link quality and fails to pass handoff call admis-sion to execute handoff in time before its received Ec/Io dropping off the requirement, Υ, a handoff forced termination occurs.
From the the above discussion, in order to design an effective dynamic cell configuration scheme that improves the system performance and minimizes the undesirable effects, it is necessary to consider the pilot power allocation and other radio resource management strategies jointly.
4.3.5 Solving DCC Problems by Reinforcement-Learning
In this chapter, we design a dynamic cell configuration (DCC) scheme that takes into account pilot, soft handoff, and maximum link power allocations as well as call admission control jointly as shown in Fig. 4.4. This scheme can be implemented in each base station being aware of the system load variation in next-generation CDMA systems. Using the proposed DCC scheme, each base station can adjust its pilot power and maximum link power constraint as well as new/handoff call admission threshold periodically. The dynamic cell configuration problem is formulated as a Markov decision process (MDP) [59]. However, traditional model-based solutions to MDP, such as policy iteration and linear programming, require a prior knowledge of the state transition probabilities and hence suffer from two
“curses”: the curse of dimensionality and the curse of modelling. The curse of dimensionality is that the complexity in these algorithms increases exponentially as the number of state increases. Dynamic cell configuration involves very large state space that makes model-based solutions infeasible. The curse of modelling is that in order to apply model-model-based
methods, it is first necessary to express state transition probabilities explicitly. In practice, this is a very difficult proposition for next-generation CDMA cellular networks due to the diverse multimedia traffic and random user mobility.
Q-learning technique was broadly adopted to solve these thorny problems in the wireless communication systems, e.g. [60] and [61]. In general, the Q-values of state-action pair are usually stored in a look-up table, but it’s impossible for the problems with continuous state spaces. Authors in [62], [63] shown that fuzzy Q-learning is an efficient technique for the approximation of continuous system states. It is an adaptation of Watkins’s Q-learning [64] for fuzzy inference systems (FIS) where both the actions and Q-functions are inferred from fuzzy rules. Taking advantage of the Q-learning technique, the universal approxima-tion property of the FIS makes the representaapproxima-tion of Q-values with large state-acapproxima-tion space possible. Moreover, a priori knowledge can be integrated in the learning procedure.
This chapter proposes a reinforcement-learning-based DCC scheme by fuzzy-Q-learning to find an optimal policy for pilot power, maximum link power constraint, soft handoff power allocation, and call admission control criterion in CDMA multimedia networks. It
This chapter proposes a reinforcement-learning-based DCC scheme by fuzzy-Q-learning to find an optimal policy for pilot power, maximum link power constraint, soft handoff power allocation, and call admission control criterion in CDMA multimedia networks. It