The wireless channel models can be generally divided into three categories as in
Fig. 2.1. The methods used in the development of those models can be sorted as
(1) Deterministic channel model, (2) Stochastic channel model, and (3) Hybrid chan- nel model (combined stochastic model and map-based model).
2.2.1
Deterministic Channel Model
The typical deterministic channel models are site-specific models as one main branch
in Fig. 2.1. It can be further resolved as (1) restored impulse responses and (2) ray
tracing models. Restored impulse responses are acquired from channel measurements, they are the most direct way to present the wireless channels. All the channel measure- ment data recorded in the literature can be considered as this type of channel model. The ray tracing models are also recorded in form of channel impulse responses, which are acquired from the deterministic solution of Maxwell’s equations or some approxi- mations based on the geographical and morphological information of certain wireless
scenarios [77]. Ray tracing models are used to complement time-consuming and ex-
pensive measurement campaigns in the investigation of propagation characteristics,
coverage, and system performance [78]–[80]. Though with the drawback that large
computational efforts are required and the results are inherently less accurate, ray tracing models have been recognised as reliable tools to simulate accurate mmWave
channel properties [81], [82]. This type of channel model method has been used as
part of standard channel models, such as IEEE 802.15.3c [83], IEEE 802.11ad [84],
MiWEBA [85], METIS [39], etc.
2.2.2
Stochastic Channel Models
Stochastic channel models, as another main branch in Fig. 2.1, have been developed
for the purpose of fast system-level simulations (compare with ray tracing models) [54].
They are the most widely used channel models in the system level simulation. They are not designed to correctly predict the impulse response in one specific location, but they
can be used to predict the probability density function (PDF) over a large area [77].
This kind of models relies on statistical observations of the channels with various
measurements in different typical scenarios. The distributions of various channel
characteristics are estimated from measurements. Stochastic channel model can be generally divided into two categories further: (1) correlation-based stochastic channel
model (CBSM or CSCM) and (2) geometry-based stochastic channel model (GBSM or GSCM).
The CBSM channel model describes the MIMO characteristics by correlation matrices. Its merit is the lower computational complexity compared with the channel models in other categories. It is widely used by system-level performance simulations. The well-known and widely used Kronecker model and Weichselberger model belong to
this category [86]. The difference between those two models is that the Kronecker
model considers the Tx and Rx are not mutually correlated, but the Weichselberger model could well present the joint correlation between Tx and Rx.
The GBSM channel models are more accurate than CBSM channel models, but they are much more computational complex. There are mainly two types of GBSM: (1) regular-shape based GBSM and (2) irregular-shape based GBSM. Most of the regular-
shape based GBSMs are analytical models, such as the Clarke model [87] and those
in [73], [88]–[91]. Typically, the scatters of the channels are distributed based on
a geometrical shape, such as ring, ellipse, etc. with certain distribution density
functions. In such an approach of modelling, all the channel parameters can be
tracked and calculated geometrically. The feature of such models is that the ref- erence models of regular-shape based GBSM are the mathematical channel models based on the geometric calculation, and the approximation methods, such as exact
Doppler, the Lp-norm method (LPNM), etc., are often used in the simulation model
to approximate the reference models. Compared with that, in the irregular-shape based GBSM, the positions of scatterers are following certain distributions and they are randomly located rather than located on a geometric shape in irregular-shape based GBSMs. The parameters used in the simulation models are estimated based on the data of real channel measurements in different scenarios. Therefore, most of
the standard channel models are in this category, such as WINNER II/+ [92], [93],
QuaDRiGa [71], COST2100 [94], etc., as well as mmWave channel models, such as
3GPP [95], METIS [39], mmMagic [29], 5GCM [1]. Note that, follow the stochstical
model branch in Fig. 2.1, the irregular-shape based GBSM can be further divided
Figure 2.2: Standard channel models and their family history.
models), (2) cluster-level approach (used in COST2100). More details can be found in [96].
Stochastic channel models are in constant development to adapt to the new features
of wireless communication systems. From the topology in Fig.2.2[54], we can see that
the leading research of channel models is following the developing line of 3GPP-SCM
(SCM stands for spatial channel model) [97], WINNER-I/II/+ [92], [93], and then
3GPP-3D [98], and 3GPP-NR [95]. Those models have covered most of the aspects of
wireless channels in various scenarios, which include network layout, the large-scale and small-scale parameters, channel scenario transition (channel segment, drop and time evolution), etc. Some Matlab codes are also available for reference from the websites of those projects. As to the other developing lines, they are not all-inclusive channel models, but they are prone to be developed for specific applications/scenarios of wireless communication, or using other approaches to better model the properties
multi-user, distributed MIMO, and moving Tx/Rx scenarios in the 4G communication
system. 1
2.2.3
Hybrid Channel Models
There is a combined stochastic and deterministic channel modelling approach, which
is called a hybrid channel model in METIS [39], as the third main branch of channel
models in Fig.2.1. Normally, in the procedure of modelling the channel, the locations
of the base stations (BSs) and the user equipment (UE) have to be fixed on a map first. The path loss (PL) and shadowing are calculated based on the map, and random shadowing objects can be generated based on ray tracing modelling method, which
explained in Section 2.2.1. Then, all other calculations are done by the stochastic
model [39]. Hybrid modelling approach takes the merits from both the deterministic
and stochastic models, and it is widely accepted in the modelling of mmWave channels.
In the mmMagic project, it extends the existing QuaDRiGa channel model [71] by
using the deterministic channel approach in the simulations [29], [99].