|
| Category | Scheme | Advantages | Disadvantages | Ref. |
|
| MAP-based algorithms | S-MAP algorithms | (i) Exploits sparsity to detect the user activity and avoid control signaling overhead
(ii) Robust to asynchronous transmissions | (i) Higher complexity
(ii) Not focused on overloaded systems | [54] |
| Approximate MAP algorithm (MMV-CS) | (i) Complexity is independent of frame length
(ii) Robust to asynchronous transmissions | (i) Additional data estimation is required
(ii) High complexity than greedy algorithms | [32] |
| Sphere decoding | (i) Maximum a posteriori performance | (i) No guarantee to terminate in polynomial time
(ii) No possibility to parallelize the computations | [24, 55, 60] |
| -best detection for sphere decoding | (i) Constant run time
(ii) Allow parallelization and pipelining | (i) Complexity increases with overloading the system
(ii) BER floor due to limited search paths | [59] |
| Greedy algorithms | OMP | (i) Lower complexity as compared to other greedy algorithms, e.g., OLS | (i) Relatively higher BER | [25] |
| OLS | (i) Lower BER than OMP | (i) Higher complexity than OMP | [25, 27] |
| GOMP | (i) Exploits block sparsity
(ii) Higher activity detection accuracy | (i) Complexity increases exponentially with group size
(ii) Performance gain depends on the frame size | [6, 26, 28] |
| IORLS | (i) Exploit block sparsity
(ii) No matrix inversion
(ii) Robust to noise | (i) The performance gain comes from large frame size, while in mMTC, data packet is typically of small size | [31] |
| SOMP (MMV-CS) | (i) Memory reduction
(ii) Faster detection
(iii) Scalable | (i) Computational complexity increases with measurement vectors | [34] |
| MMP (MMV-CS) | (i) Complexity is independent of frame length
(ii) Robust to asynchronous transmissions | (i) Additional data estimation step
(ii) Simulation parameters are not realistic for mMTC | [32] |
|
