Sunteți pe pagina 1din 3

Revisor 1.

4656329.v1 Review Report


Subject Appropriateness of the Manuscript
The topic of this manuscript falls within the scope of Wireless Communications and
Mobile Computing
Recommendation
Consider after Major Changes
Comments
This paper presents a Markov Mode-Multiplexing LINC Architecture in OFDM
Transmitters.
The paper is well written and theory, simulation and experimental results are well
balanced.
The presented results are interesting but I think the authors should also discuss the
applicability of this technique to another type of signals.
Please include actual efficiency results and discuss if the presented values include or
not the combiner effect.
Also present efficiency results for the simulations.
It would be helpful if the authors could compare their LINC results with a classical
class B PA. This would help to justify (or not) the importance of their work.
Revisor 2

4656329.v1 Review Report


Subject Appropriateness of the Manuscript
The topic of this manuscript falls within the scope of Wireless Communications and
Mobile Computing
Recommendation
Consider after Major Changes
Comments
The manuscript reports a variant to the Mode-Multiplexing LINC (MM-LINC)
architecture for power amplifiers (PA), targeting for simultaneous high linearity and
efficiency. The proposed technique, named Markov Mode-Multiplexing LINC
architecture (or simply Markov-LINC) consists of defining a different way to change
between the outphasing and balanced modes of LINC. In the MM-LINC, this selection
is based on the amplitude of the input signal (compared to a predefined threshold
value), while in the proposed Markov-LINC it is independent of the values of the input
signal samples, and only depends on a binary sequence that is generated by a Markov
chain process but whose statistical parameters are a priori determined from the input
signal statistics.

In my opinion, I recognize the relevance of considering the information that is within


the signal statistics, as an alternative to considering the values of the isolated samples.
However, this work presents significant issues that do not allow me to check if the
proposed method is relevant or not, when compared to MM-LINC.
First, it is not clear if the resulting efficiency and linearity are very sensitive, or not, to
the values selected for the a and b parameters (in eq. (8)). As the Authors retrieve
these values from the statistical properties of a specific signal realization, and since the
PA (simulated or measured) is then tested with that same signal realization, no real
cross-validation is performed. It could be the case that, with a different signal
realization (as would happen in practice), for which a and b would be different
from those previously determined, the PA efficiency and linearity might not show such
good results.
Secondly, it is not sufficiently clear how the threshold parameter of the MM-LINC
method was determined/optimized. On page 4 (2nd column) it is indicated that it is
taken to be equal to the average amplitude of the input signal, but is this the best case?
Thirdly, the true advantages of the Markov-LINC over the MM-LINC are not
completely demonstrated, as it is not clear to me that not considering the input samples
(for deciding if outphasing or balanced mode is to be set) is, in fact, more effective or
efficient (from an implementation view point) that simply comparing such samples
against a threshold (which is computationally very simple). Also, the performance
values obtained in Table III are only slightly different from the MM-LINC results
(notice, for instance, in Table III, that the Markov-LINC increases the P_DC, with
respect to MM-LINC, for the 1.4 MHz signal; while it decreases the P_DC for the 5
MHz signal thus, it is not showing consistency how can the Reader be convinced
that the observed improvements were not a coincidence?). In addition, it seems that the
ACPR results shown in Table III do not corroborate the spectra shown in Figs. 8
(especially Fig. 8.a), where a clear improvement is seen for LINC and Markov-LINC
over MM-LINC, but only 1.4 dB of improvement is presented in Table III).
Finally, it is not clear the selection used for the a and b values. On page 4 (2nd
column), it is indicated that these were both considered equal to 0.99. Was this the
value used in the simulated and measured results of Sections 4 and 5? Moreover,
considering a value so close to 1 means that the outphasing and balanced modes are
practically alternating on a sample-by-sample basis. What would be the result that
would be achieved if a and b are, in fact, set to 1 (and the mode control signal is a
simply alternating binary signal)? If the results would be similar to those presented in
Sections 4 and 5, would it still be meaningful to involve the Markov process statistics
issues into consideration?

Beyond the lack of proper validation that I have mentioned above, I would also like to
point out additional issues in specific parts of the manuscript, as I present in the
following enumerated list.

1. Page 2, 1st column, 6th line from the end:


Instead of sampling time use sampling period, since sampling time is more
commonly referred to the time instant on which a sample was taken.

2. Page 2, eq. (6):


For the case of pi(n)=1, shouldnt the matrix be [0.5 0; 0 0.5] instead of [1 0; 0 1], for
maintaining the power of the input signal?

3. Page 3, 1st column, line 7:


Shouldnt it be state value instead of state space? Space is a set of values.

4. Page 4, 2nd column, eq. (12):


It is not straightforward how eq. (12) is reached from solving (8) with (9). This
should be better explained in the manuscript.

5. Page 5, eq. (30) and the text below it:


The Authors mention the use of a Memory Polynomial (MP) model for mimicking the
PA behavior, however it is claimed that no memory is considered in this model (as
observed in Figs. 4.a) and 5.a)). Therefore, only a standard static polynomial is being
used. I think it would be much more appropriate to state from the start that the used
model is that of a static polynomial, instead of introducing the MP model which has
features that are not used by the Authors.

6. Page 6, 1st column, first sentences:


It is not clear to me how the Authors reach the conclusion, from Figs. 4.b) and 5.b),
that the best results are achieved for a approximately equal to b. Nor that a better
ACPR is reached for b>>a, and a worse ACPR is reached for a>>b.

7. Page 7, Figs. 6.a) and 6.b):


Why is the spectrum corresponding to the LINC case so perfect (it doesnt even have
the noisy aspect, which the other spectra have)?

S-ar putea să vă placă și