Page 138 - ITU Journal Future and evolving technologies – Volume 2 (2021), Issue 2

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 2

o
digital BF ilters. Similar simpli ication was gained by with a tap exhibiting 0.065 phase and 0.01dB magnitude

the multi‑AUT‑TX canceler hardware, where the reduced errors, respectively, results in approximately 60dB of SI

number of AUX TX RF chains were jointly designed with cancellation. Hence, the considered multi‑tap canceler

MUXs/DEMUXs and TX/RX digital BF. We have presented architecture in our simulations is capable of delivering

a general optimization framework for the joint design of approximately 60dB of analog cancellation per tap.
analog SI cancellation and digital BF, and detailed a spe‑
ci ic algorithmic solution targeting the FD rate maximiza‑ A.2 Model for the AUX TX RF chains

tion. The performance evaluation results, based on the

proposed realistic models for non‑ideal analog cancella‑ One known characteristic of the canceler architecture
tion hardware, demonstrated that the presented designs based on AUX TX RF chains is that the SI signal used for
can be implemented with less cancellation elements (less cancellation at the RX side is obtained from the digital do‑

taps or AUX TXs) than SotA ones, while achieving larger main. Due to this fact, this cancellation signal does not

FD rates. We discussed recent STAR schemes capitaliz‑ include the inherit non‑linearities of the actually trans‑
ing on the overviewed hardware and algorithmic frame‑ mitted SI signal; these non‑linearities exist in real‑world

work, and presented some open challenges and research TX RF chain hardware. As has been described in [54],

directions for future FD MIMO communication systems one such non‑linearity is the oscillator phase noise at the

and their promising wireless applications. TX RF chains and AUX TX RF chains. This non‑linearity
source has been shown to be a dominant bottleneck for
A. SIMULATION MODELS FOR THE ANALOG the performance of analog cancelers based on AUX TX RF
CANCELER HARDWARE chains. Thus, our model includes phase noise effects.

Let us denote by and with = 1, 2, … , and

We next present two simulation models for non‑ideal ana‑ = 1, 2, … , the phase noise due to the ‑th AUX TX RF

log canceler hardware. The irst model is considered for chain and ‑th TX RF chain, respectively. We use the ma‑

the presented multi‑tap canceler architecture and the sec‑ trix notation ∈ ℂ × to represent the imperfections

ond for the multi‑AUX‑TX canceler architecture. due to phase noise. Each ( , )‑th element of this matrix

captures the phase noise mismatch between the ‑th AUX
A.1 Model for the analog taps TX RF chain and ‑th TX RF chain, and is expressed as:
In the ideal hardware case, the amplitude and phase of [Φ ] = exp ( ) − exp ( ) + 1, (18)

each of the analog taps take any desired arbitrary value. ,
However, the settings for the attenuator and phase shifter for = 1, 2, … , and = 1, 2, … , . In our simula‑

comprising a tap take only discrete value steps when re‑ tions we do not use the ideal cancellation values given
alistic hardware is considered. Consequently, we assume by matrix C , we instead use a more realistic noisy ver‑

that each tap is set with steps of 0.02dB for attenuation sion given by C , which is computed as C = L L , where
̂
̂
̂
5 4
o
and of 0.13 for phase; these values match the step val‑ the matrix L ̂ ∈ ℂ × is de ined as L ̂ ≜ ⊙ L .
ues reported in [15]. Thus, for each analog tap in our 4 4 4
simulations, the phase setting has a random phase error Notice than in the ideal case of zero phase noise (i.e.,

o

o
uniformly distributed between −0.065 and 0.065 , and = = 0), has all its entries equal to one,
̂
̂
the amplitude setting has a random amplitude error uni‑ hence, L and C become equal to L and C , respectively.

4
4
formly distributed between −0.01dB and 0.01dB. More We model the phase noises as zero‑mean normal random
2
2
speci ically, in our simulations we do not use the ideal variableseachwithvariance , andwesetas thevalue

o
cancellation values given by C , instead we use a more re‑ of the phase noise jitter0.717 , as computed in [54] for the

̂
alistic noisy version given by C ∈ ℂ × . The non‑ MAX2829 oscillator. Note that this value has been used

̂
zero elements of C are the same non‑zero elements of in several FD experiments using the analog canceler ar‑

chitecture based on multiple AUX TXs [14, 31]. We also
C , but affected by a random phase and magnitude error. assume that exp( ) and exp( ) ∀ = 1, 2, … , and

More speci ically, for the ( , )‑th non‑zero element of C ∀ = 1, 2, … , are IID random variables; this implies

due to the ‑th analog tap, we compute its noisy version that our model considers the case where the TX RF chains
as follows:
have independent oscillators. As was discussed in [54],
̂

[C ] = [C ] exp ( ) 10 /20 (17) the amount of analog cancellation for these types of phase
,
,

noise is approximately 35dB. This means that the consid‑
for = 1, 2, … , and = 1, 2, … , , where ered multi‑AUX‑TX canceler architecture in our simula‑

is uniformly distributed over the interval tions is capable of delivering approximately 35dB of ana‑
[−0.065 /180, 0.065 /180] and is uniformly dis‑ log cancellation per AUX TX RF chain.
tributed over the interval [−0.01, 0.01]. In the latter ex‑
pression, and represent the phase and magnitude REFERENCES

errors, respectively, due to the non‑ideal hardware at the
‑th tap. We also assume that and ∀ = 1, 2, … , [1] “The Next Hyper‑Connected Experience for All,”

are IID random variables. Applying analog cancellation White Paper, Samsung 6G Vision, Jun. 2020.
124 © International Telecommunication Union, 2021