Mobility in e-UTRAN is based on hard handover rather than on soft handover as in UMTS [11]. The mechanism of hard handover has been used in 2nd generation GSM networks and has shown to be efficient for mobility management. In a hard handover, the user keeps the connection to only one cell at a time, breaking the connection with the former cell immediately before making the new connection to the target cell. The basic concept of handover as in GSM is likely to be implemented in e-UTRAN except for the handover preparation phase, which requires new mechanisms.
The reason for abandoning soft handover is related to the extra-complexity involved in its implementation, and the fact that it is not suitable for inter-frequency handover. Furthermore, as explained in the previous chapter, soft handover handicaps system capacity in highly loaded network condition and with high number of users in soft handover situation. To guarantee seamless and lossless hard handover in e-UTRAN, the handover triggering time should be as low as possible [98].
In general, there are different causes for handover: a user can move to another cell because he verifies a power budget condition or because he experiences bad channel quality. In this work we consider only the first case, namely power budget based handover. This handover is based on the comparison of the received signal strength from the serving cell and from the neighbouring cells.
5.4.1 E-UTRAN handover algorithm
In order to study the performance of the auto-tuning of e-UTRAN hard handover, some assumptions for the call admission control (CAC) and resource allocation are made. In the CAC algorithm, a user can be admitted to the network only when the following conditions are fulfilled: • Good signal strength: the mobile selects the cell that offers the maximum signal. If this signal is lower than a specified threshold then the mobile is blocked because of coverage shortage. This condition is in fact a selection criterion. It is noted that in 3GPP, there is no specification for LTE cell selection and reselection.
• Resource availability in the selected cell: the mobile can be granted physical resources in terms of resource blocks between a minimum and maximum threshold. When the signal strength condition is satisfied, the eNB checks for resource availability. If the available resource is lower than a minimum threshold, the call is blocked.
Hard handover is performed in this study using a similar algorithm to the one used in GSM: while in communication, the mobile periodically measures the received power from its serving eNB and from the neighbouring eNBs. The mobile, initially connected to a cell k, triggers a handover to a new cell i if the following conditions are satisfied:
• The Power Budget Quantity (PBQ) is higher than the handover margin:
( )k
i
Hysteresis
HM
P
P
PBQ=
i*−
k*≥
,
+
(5.6)where Pk* is the received power from the eNB k expressed in dB; HM(k,i) is the handover margin between eNB k and i; the Hysteresis is a constant independent of the eNBs and mobile stations and is fixed in this study to 0.
• The received power from the target eNB must be higher than a threshold. This is the same condition as in the CAC process.
• Enough resource blocks are available in the target eNB.
The last condition requires information exchange between eNBs because the original cell has to know a priori the load of the target cell; otherwise the handover is blind and the communication risks to be dropped. In an inter-eNB handover procedure, the source eNB is responsible for performing handover preparation to the target eNB based on measurement report transmitted by the mobile. If the highest ranked cell listed in the mobile measurement report is congested and can not admit incoming calls especially real-time service calls, the source eNB performs handover preparation to the cell with the next highest rank. On the other hand, if the source eNB has the load knowledge of its neighbour's eNBs, it could efficiently decide to which cell the handover should be performed before initiating handover preparation procedures [107]. Therefore, the source eNB needs to perform handover decision considering load information. In 3GPP proposals, there are mainly two solutions for the source eNB to know whether the highest ranked cell can admit the incoming call or not. The first solution is to standardize load
information exchange on the interface X2 as a common measurement, whereas the second solution concerns the implication of the target cell in the handover preparation phase. The target cell can reply a handover preparation failure for example if it is loaded.
The first solution achieves shorter delay in handover preparation phase; however it requires the definition of a new eNB measurement for load information. The exact definition of eNB load is still under discussions in 3GPP. The second solution seams to be similar to the existing handover procedures (in UMTS and GSM) where base stations are not aware of the load status of their neighbours. In this second solution, the target eNB has to respond to each handover request message regardless of its congestion state. As a result, the processor load and signaling messages in the X2 interface can rise. If the highest ranked cell can not accept the call, the source eNB is compelled to try handover preparation to another eNB resulting in a longer delay for handover preparation phase. The first solution is preferred to the second one because of the requirement that the hard handover be lossless and the handover triggering time should be as low as possible.
5.4.2 Handover adaptation and load balancing
From the previously described handover algorithm, a low handover margin allows users to be connected to the closest cell everywhere in the cell, but ping-pong effect may occur frequently. On the other hand, a high handover margin generates high interference for cell-edge users especially with the use of low frequency reuse factor, but ping pong effect problems are avoided. Adapting the handover margin allows to achieve interesting trade-offs between different network states.
Figure 5.4 presents a typical handover situation, whereas Figure 5.5 shows a situation where handover thresholds are adapted to the relative cell load, rather than being constant. In Figure 5.5 Some of users in the handover zone that would otherwise be served by the congested cell (cell k) are now handed over to the less congested cell. This can be achieved by delaying the handover to the congested cell and advancing the handover from the congested cell or in other words decreasing the handover margin from cell k to cell i and increasing the handover margin for the other direction. In fact, decreasing handover margin from cell k to cell i allows more users to verify the power budget handover condition. So, it leads to a decrease of the service area of congested cells and to an increase of the service area of less-loaded cells.
Auto-tuning in a non-uniformly loaded network could be particularly beneficial in e-UTRAN. It implements a simple load balancing mechanism which increases the overall capacity of the system, by simply distributing the load more evenly between the neighbouring cells.
Figure 5.4. Typical pattern of geographical distribution of HO procedure.
Figure 5.5. Example of geographical distribution of HO procedure with traffic balancing.
5.4.3 Auto-tuning of handover margin
The auto-tuning aims at dynamically adapting handover margins between cells as a function of their loads, to optimize network performance. Each coefficient of the matrix HM governs the traffic flows between two cells. The coefficient HM(k,i) depends only on the difference between the load of cell i and k. Define the handover margin matrix HM as
( )k
i
f
(
k i)
HM
,
=
χ
−χ
(5.7)The function f should satisfy:
(i) f is a decreasing function from the interval [-1,1] to [HMmin, HMmax],
(ii)
f( ) ( )x
+
f
−x
=2f( )0,∀x∈[−1,1]
,where HMmin and HMmax are respectively the minimum and the maximum values of the handover
margin. f(0) is the value of the planned handover margin since the planning process assumes the uniformity of the cell loads.
Handover from k to i Handover from i to k
Cell k
Cell i
Handover from k to i Handover from i to kCell k
Cell i
The first condition implies that when the cell k is fully loaded and i does not serve any mobile, (i.e. χk -χi approaches 1) it is worth keeping the handover margin HM(k,i) to the lowest value.
For the second condition, let x be defined as the difference between loads of cell k and i (i.e. x=χk
-χi); this condition is used to avoid ping pong effect. It implies that when the cell k is over loaded
and the cell i is less-loaded, cell k pushes mobiles to cell i and conversely, cell i delays handover to cell k.
The function f can be approximated by a polynomial (with Taylor series expansion). The polynomial coefficients can be dynamically determined using learning techniques. The concept of fuzzy reinforcement learning presented earlier in previous chapters could be well suited. In the present study, we restrict the development of f(x) to the order 1:
( )x
f( )
(f( )
HM
)x
f
=
0
+
0
−
max (5.8)The development of order 0 corresponds to the classical case without any auto-tuning. The simulations aim at comparing the development of order 1, namely with auto-tuning to the case without auto-tuning.