Radio calibration against 3GPP (TR 38.901 clause 7.8.1)

Summary (1.8.0). The radio model of HandoverLab -- path loss, LOS probability, shadowing and building penetration loss of TR 38.901 -- passes the large-scale calibration that 3GPP defined for TR 38.901. In the calibration drop of clause 7.8.1, the CDFs of the coupling loss and of the geometry (wideband SINR, with and without noise) lie inside the range of the 21 companies that reported their results to 3GPP in R1-165974, for UMa, UMi-Street Canyon and the indoor office at 6 GHz and 30 GHz: 325 of 342 compared percentiles (5 % to 95 % in steps of 5 %) are inside the range, and none is more than about 1 dB from the industry mean. handoverlab verify repeats the check (group calibration). With c = 3.0e8 m/s in the breakpoint distances (2.0.0) every number of this page is unchanged.

Version. The drop uses the formulas of TR 38.901 V19.5.0 with one exception: the IRR-glass loss of the high-loss O2I model. V19.5.0 Table 7.4.3-1 gives 25.4 + 0.11 f dB (note 2: updated in the Release 19 study on 7-24 GHz); the calibration keeps the earlier 23 + 0.3 f dB of TR 38.901 V17.0.0, because the companies of R1-165974 worked before that change (IRR_GLASS_R1_165974 in handoverlab.verify.calibration; pass irr_glass=IRR_GLASS_V19 to simulate_drops for the V19.5.0 value). With the V19.5.0 value (30 000 users, seed 1) the 6 GHz cases stay inside the company range, but at 30 GHz the coupling loss and the SINR of UMa and UMi leave it (coupling loss: UMa 15 of 19 percentiles inside, largest distance from the mean 3.30 dB; UMi 11 of 19, 3.25 dB): the higher glass loss at 30 GHz is a change of the model, not an error of the reproduction.

HandoverLab against the 21 companies of 3GPP R1-165974

The reference

TR 38.901 clause 7.8.1 fixes the large-scale calibration (Table 7.8-1) and points to the 3GPP RAN1 contribution R1-165974, "E-mail discussion summary of the large scale calibration" (RAN1 #85, Nanjing, May 2016; 3GPP server). Its workbook holds the CDFs of 21 companies (Huawei, CATT, ITRI, LGE, InterDigital, CHTTL, Samsung, Intel, Sharp, CATR, ZTE, Ericsson, NTT DOCOMO, Nokia, Qualcomm, Xinwei, CMCC, Lenovo, KDDI, ETRI, NEC; 20 for the indoor office) at every percentile. src/handoverlab/verify/data/r1_165974.csv keeps, per case, metric and percentile, the mean over the companies, the smallest and the largest company value and the number of companies. The companies produced the results with TR 38.900 V14.0.0; TR 38.901 still names R1-165974 as the calibration results of clause 7.8.1.

The calibration drop

handoverlab.verify.calibration reproduces the drop of Table 7.8-1 with the propagation formulas of the simulator (a test checks that they give exactly the values of handoverlab.channel.pathloss):

Item Setting Source
Layout UMa: 19 sites, ISD 500 m, BS 25 m; UMi: 19 sites, ISD 200 m, BS 10 m; geographical wrap-around TR 38.901 Tables 7.2-1 and 7.8-1
Indoor office 12 ceiling sites, 20 m apart, in a 120 m x 50 m hall, BS 3 m, users 1 m Table 7.2-2, Figure 7.2-1
Sectors and antenna 3 sectors per site (30, 150, 270 deg); element pattern of Table 7.3-1 (8 dBi); 10 vertical elements, d_V = lambda / 2, DFT tilt 102 deg (UMa, UMi), 110 deg (indoor) Tables 7.3-1, 7.8-1; Eq. 7.3-1
Users uniform; UMa and UMi 80 % indoor at height 3 (n_fl - 1) + 1.5 m, n_fl ~ uniform(1, N_fl), N_fl ~ uniform(4, 8); minimum 2D distance 35 m (UMa), 10 m (UMi), applied to d2D-out Table 7.2-1 ("Same as 3D-UMa in TR36.873"); the formula as in ITU-R M.2412-0 Table 5b (TR 36.873 itself is not in the local sources)
Path loss and LOS Table 7.4.1-1 (UMa with the random effective environment height of note 1), Table 7.4.2-1 (UMa with C'(h_UT), indoor open office) TR 38.901
Shadowing LOS / NLOS of Table 7.4.1-1; 7 dB for indoor users of UMa and UMi; independent between sites, shared by the sectors of a site Tables 7.4.1-1 and 7.5-6
O2I loss 50 % low loss, 50 % high loss; d2D-in = min of two uniform(0, 25 m); IRR glass 23 + 0.3 f dB (see Version above) clause 7.4.3.1, Tables 7.4.3-1 (V17.0.0 value), 7.4.3-2
Link budget 49 dBm (UMa), 44 dBm (UMi) at 6 GHz with 20 MHz; 35 dBm at 30 GHz with 100 MHz; 24 dBm indoor; UT noise figure 9 dB; isotropic UT Table 7.8-1
Association sector with the highest received power (handover margin 0 dB), no fast fading Table 7.8-1

Results

30 000 users per case, seed 1 (handoverlab verify --full):

Case Metric Median: HandoverLab / 3GPP mean [dB] Percentiles 5-95 % inside the company range Largest distance from the mean [dB]
UMa-6 GHz coupling loss 119.6 / 120.3 19 / 19 0.76
UMa-6 GHz SINR 5.6 / 5.8 19 / 19 0.39
UMa-6 GHz SIR 8.0 / 8.5 16 / 19 0.79
UMa-30 GHz coupling loss 139.2 / 140.1 19 / 19 1.02
UMa-30 GHz SINR -19.2 / -19.8 19 / 19 0.73
UMa-30 GHz SIR 7.8 / 8.6 10 / 19 0.80
UMi-6 GHz coupling loss 111.5 / 112.2 19 / 19 0.90
UMi-6 GHz SINR 5.3 / 5.5 19 / 19 0.61
UMi-6 GHz SIR 6.8 / 7.3 19 / 19 0.76
UMi-30 GHz coupling loss 131.7 / 132.4 19 / 19 0.83
UMi-30 GHz SINR -11.8 / -12.2 19 / 19 0.53
UMi-30 GHz SIR 6.9 / 7.5 14 / 19 0.77
Indoor-6 GHz coupling loss 59.4 / 59.3 19 / 19 0.20
Indoor-6 GHz SINR 3.0 / 3.0 19 / 19 0.16
Indoor-6 GHz SIR 3.0 / 3.0 19 / 19 0.16
Indoor-30 GHz coupling loss 73.3 / 73.3 19 / 19 0.20
Indoor-30 GHz SINR 3.0 / 3.1 19 / 19 0.13
Indoor-30 GHz SIR 3.0 / 3.1 19 / 19 0.13

The 17 percentiles outside the range are all SIR values of UMa and UMi in the lower half of the distribution, at most 0.12 dB below the smallest company value: the SIR of HandoverLab is about 0.5 dB below the industry mean there, which is on the conservative side. handoverlab verify passes a case when every percentile lies inside the company range widened by 0.5 dB.

What this calibration covers, and what it does not

  • It covers the radio model: the path-loss formulas, the LOS probabilities, the shadowing, the O2I loss and the frequency dependence, with the deployment and antenna that 3GPP fixed for calibration. These are the same formulas the handover engine uses.
  • The handover engine uses its own antenna abstraction (one beam gain, radio.bs_gain_db, and an isolation factor for interfering beams) instead of the sectorised array of Table 7.8-1, and a spatially consistent, moving radio map instead of independent drops. Received powers in a handover scenario therefore differ from the calibration drop by the antenna model, not by the propagation model.
  • The failure model (link monitoring, T310, the outcome of a handover) is not part of any 3GPP calibration; its comparison with ns-3 is on the page failure-model calibration. Read failure rates relative to each other.

Repeat it

handoverlab verify --only calibration          # 10 000 users per case, a few seconds
handoverlab verify --only calibration --full   # 30 000 users per case, the table above
python tools/make_calibration_figure.py        # docs/images/calibration_38901.png
from handoverlab.verify import calibration as cal

case = cal.CalScenario("UMa", 6.0)
res = cal.simulate_drops(case, n_users=30000, seed=1)
for c in cal.compare(case, res):
    print(c.metric, c.inside.sum(), "of", len(c.percentiles), "inside, largest distance", c.max_dev_db)