Harald Welte | a5312fd | 2009-11-29 17:41:22 +0100 | [diff] [blame] | 1 | Ideas about a handover algorithm |
| 2 | ====================================================================== |
| 3 | |
| 4 | This is mostly based on the results presented in Chapter 8 of "Performance |
| 5 | Enhancements in a Frequency Hopping GSM Network" by Thomas Toftegaard Nielsen |
| 6 | and Joeroen Wigard. |
| 7 | |
| 8 | |
| 9 | === Reasons for performing handover === |
| 10 | |
| 11 | Section 2.1.1: Handover used in their CAPACITY simulation: |
| 12 | |
| 13 | 1) Interference Handover |
| 14 | |
| 15 | Average RXLEV is satisfactory high, but average RXQUAL too low indicates |
| 16 | interference to the channel. Handover should be made. |
| 17 | |
| 18 | 2) Bad Quality |
| 19 | |
| 20 | Averaged RXQUAL is lower than a threshold |
| 21 | |
| 22 | 3) Low Level / Signal Strength |
| 23 | |
| 24 | Average RXLEV is lower than a threshold |
| 25 | |
| 26 | 4) Distance Handover |
| 27 | |
| 28 | MS is too far away from a cell (measured by TA) |
| 29 | |
| 30 | 5) Power budget / Better Cell |
| 31 | |
| 32 | RX Level of neighbor cell is at least "HO Margin dB" dB better than the |
| 33 | current serving cell. |
| 34 | |
| 35 | === Ideal parameters for HO algorithm === |
| 36 | |
| 37 | Chapter 8, Section 2.2, Table 24: |
| 38 | |
| 39 | Window RXLEV averaging: 10 SACCH frames (no weighting) |
| 40 | Window RXQUAL averaging: 1 SACCH frame (no averaging) |
| 41 | Level Threashold: 1 of the last 1 AV-RXLEV values < -110dBm |
| 42 | Quality Threshold: 3 of the last 4 AV-RXQUAL values >= 5 |
| 43 | Interference Threshold: 1 of the last AV-RXLEV > -85 dBm & |
| 44 | 3 of the last 4 AV-RXQUAL values >= 5 |
| 45 | Power Budget: Level of neighbor cell > 3 dB better |
| 46 | Power Budget Interval: Every 6 SACCH frames (6 seconds ?!?) |
| 47 | Distance Handover: Disabled |
| 48 | Evaluation rule 1: RXLEV of the candidate cell a tleast -104 dBm |
| 49 | Evaluation rule 2: Level of candidate cell > 3dB better own cell |
| 50 | Timer Successful HO: 5 SACCH frames |
| 51 | Timer Unsuccessful HO: 1 SACCH frame |
| 52 | |
| 53 | In a non-frequency hopping case, RXQUAL threshold can be decreased to |
| 54 | RXLEV >= 4 |
| 55 | |
| 56 | When frequency hopping is enabled, the following additional parameters |
| 57 | should be introduced: |
| 58 | |
| 59 | * No intra-cell handover |
| 60 | * Use a HO Margin of 2dB |
| 61 | |
| 62 | === Handover Channel Reservation === |
| 63 | |
| 64 | In loaded network, each cell should reserve some channels for handovers, |
| 65 | rather than using all of them for new call establishment. This reduces the |
| 66 | need to drop calls due to failing handovers, at the expense of failing new call |
| 67 | attempts. |
| 68 | |
| 69 | === Dynamic HO Margin === |
| 70 | |
| 71 | The handover margin (hysteresis) should depend on the RXQUAL. Optimal results |
| 72 | were achieved with the following settings: |
| 73 | * RXQUAL <= 4: 9 dB |
| 74 | * RXQUAL == 5: 6 dB |
| 75 | * RXQUAL >= 6: 1 dB |
| 76 | |
| 77 | |
Harald Welte | 8c83af6 | 2009-11-29 20:02:53 +0100 | [diff] [blame] | 78 | |
| 79 | == Actual Handover on a protocol level == |
| 80 | |
| 81 | After the BSC has decided a handover shall be done, it has to |
| 82 | |
| 83 | # allocate a channel at the new BTS |
| 84 | # allocate a handover reference |
| 85 | # activate the channel on the BTS side using RSL CHANNEL ACTIVATION, |
| 86 | indicating the HO reference |
| 87 | # BTS responds with CHAN ACT ACK, including GSM frame number |
| 88 | # BSC sends 04.08 HO CMD to MS using old BTS |
| 89 | |