Implementation of LOS TDL-D / TDL-E Channels in srsRAN_4G

Date: 2026-03-04
Repository: srsRAN_TDL_LOS_JOAO/srsRAN_4G
Version: 1.0

---

1. Objective

Add full support for the TDL-D and TDL-E channel models (LOS channels as per 3GPP TR 38.901) to the internal channel emulator in srsRAN_4G, while maintaining full backwards compatibility with the existing NLOS models (EPA/EVA/ETU/TDL-A/B/C).

---

2. Context and Motivation

The TDL-D and TDL-E models represent propagation scenarios with Line of Sight (LOS) between transmitter and receiver. Unlike the NLOS models (TDL-A/B/C), under LOS the first tap follows a Rician distribution (mixture of a deterministic component + a diffuse Rayleigh component), while the remaining taps retain the usual Rayleigh behaviour.

2.1 Tap 0 Equation (Rice)

$$a_0(t) = \sqrt{P_0} \left( \sqrt{\frac{K}{K+1}} \cdot e^{j(2\pi f_S t + \phi_0)} + \sqrt{\frac{1}{K+1}} \cdot g_0(t) \right)$$

• K: linear K-factor of the profile (TDL-D: 13.3 dB = 21.38; TDL-E: 22.0 dB = 158.49)
• f_S = 0.7 · f_D: Doppler frequency of the LOS peak (as per TR 38.901)
• φ_0: initial phase uniformly distributed in [0, 2π), randomly generated at start-up
• g_0(t): Rayleigh process (Jakes SoS, already present in the engine)
• P_0 = 0 dB: normalised total power of tap 0 (as per the 3GPP table note)

2.2 Taps 1..L (Rayleigh, same as NLOS models)

$$a_n(t) = \sqrt{P_n} \cdot g_n(t), \quad n \geq 1$$

---

3. Parameter Tables (DS = 50 ns)

TDL-D — Tap 0 Rice + remaining Rayleigh taps (total 13 indices, tap[0] = Rice)

Idx	Delay norm	Delay (ns)	Power (dB)	Type
0	0.000	0.00	0.0 total	Rice (K=13.3 dB)
1	0.035	1.75	-18.8	Rayleigh
2	0.612	30.60	-21.0	Rayleigh
3	1.363	68.15	-22.8	Rayleigh
4	1.405	70.25	-17.9	Rayleigh
5	1.804	90.20	-20.1	Rayleigh
6	2.596	129.80	-21.9	Rayleigh
7	1.775	88.75	-22.9	Rayleigh
8	4.042	202.10	-27.8	Rayleigh
9	7.937	396.85	-23.6	Rayleigh
10	9.424	471.20	-24.8	Rayleigh
11	9.708	485.40	-30.0	Rayleigh
12	12.525	626.25	-27.7	Rayleigh

Note: Tap 0 has K₁ = 13.3 dB and an average power of 0 dB.

TDL-E — Tap 0 Rice + remaining Rayleigh taps (total 14 indices, tap[0] = Rice)

Idx	Delay norm	Delay (ns)	Power (dB)	Type
0	0.0000	0.00	0.0 total	Rice (K=22.0 dB)
1	0.5133	25.67	-15.8	Rayleigh
2	0.5440	27.20	-18.1	Rayleigh
3	0.5630	28.15	-19.8	Rayleigh
4	0.5440	27.20	-22.9	Rayleigh
5	0.7112	35.56	-22.4	Rayleigh
6	1.9092	95.46	-18.6	Rayleigh
7	1.9293	96.47	-20.8	Rayleigh
8	1.9589	97.95	-22.6	Rayleigh
9	2.6426	132.13	-22.3	Rayleigh
10	3.7136	185.68	-25.6	Rayleigh
11	5.4524	272.62	-20.2	Rayleigh
12	12.0034	600.17	-29.8	Rayleigh
13	20.6519	1032.60	-29.2	Rayleigh

Note: Tap 0 has K₁ = 22.0 dB and an average power of 0 dB.

---

4. Modified Files

Only 2 files in the fading engine are modified. The rest of the stack requires no changes thanks to the existing architecture.

4.1 lib/include/srsran/phy/channel/fading.h

Changes:

1. Add srsran_channel_fading_model_tdld and srsran_channel_fading_model_tdle to the enum.
2. Add 3 fields to the srsran_channel_fading_t struct for the LOS state:
   • bool is_los — enables the Rice branch on tap 0
   • float k_factor — linear K for tap 0
   • float los_phi0 — random initial phase φ₀

4.2 lib/src/phy/channel/fading.c

Changes:

1. Extend nof_taps[9] with 2 entries (TDL-D: 13, TDL-E: 14).
2. Extend excess_tap_delay_ns[9][...] with the TDL-D/E tables (DS=50 ns).
3. Extend relative_power_db[9][...] with the Rayleigh tap powers. To ensure a total mean gain of 1.0 (0 dB), a normalisation offset has been applied to each table:
   • TDL-D: -0.1446 dB offset applied to all taps.
   • TDL-E: -0.1707 dB offset applied to all taps.
4. Add parse_model() entries for tdld/tdl-d and tdle/tdl-e.
5. Add helper get_los_phasor(t, f_d, phi0) → cf_t.
6. Modify generate_taps(): for q->is_los at tap i==0, combine scaled LOS + Rayleigh components.
7. Modify srsran_channel_fading_init(): detect LOS model, initialise is_los, k_factor, los_phi0.

---

5. Solution Architecture (Maximum Reuse)

fading_execute() └─► generate_taps(q, time) │ ├─ tap i=0, is_los=true: │ rayleigh_coeff = get_doppler_dispersion(...) ← REUSED │ los_coeff = get_los_phasor(t, 0.7*fd, phi0) ← NEW │ a = sqrt(P0) * (sqrt(K/(K+1))*los_coeff + sqrt(1/(K+1))*rayleigh_coeff) │ └─ tap i>0 (or is_los=false for NLOS): a = get_doppler_dispersion(...) ← UNCHANGED

The whole FFT/Overlap-Add pipeline in filter_segment() is reused without changes.

---

6. Compatibility

Model	Status
none	✅ Unchanged
epaX, evaX, etuX	✅ Unchanged
tdlaX, tdlbX, tdlcX	✅ Unchanged
tdldX	🆕 TDL-D LOS (Rice tap 0, K=13.3 dB)
tdleX	🆕 TDL-E LOS (Rice tap 0, K=22.0 dB)

Usage in ue.conf:

[channel.ul.fading]
enable = true
model  = tdld3       ; TDL-D with 3 Hz Doppler
; model = tdle5      ; TDL-E with 5 Hz Doppler

---

7. Verification Strategy

• Clean build with make srsue -j$(nproc) from the build/ directory.
• Unit test fading_channel_test with tdld and tdle models.
• NLOS regression with tdla, epa to confirm backwards compatibility.
• Integration in an srsUE session with LOS fading enabled.

See validation.md for the full results.