[STDS-802-16] [PREAMBLE] Some kick-off discussion for Aug 10 CC
Dear preamble ad-hoc participants,
I would like to initiate some technical discussion to prepare for
tomorrow's conference call. I am supposed to provide comparison results
on different sequences, but I find it is difficult to do so and perhaps
not that meaningful at this stage if we still cannot agree on the
functions we want to achieve. So, the functional requirements are still
the focus here.
The purpose of the discussion is to help (me at least) to better
understand where the issues are, whether the existing preamble is
adequate, and whether the proposed modification is worthwhile. I hope
people can fill in with what I do not know and what I stated
incorrectly, and thus drive the same goal together, which is to make the
system better.
1) Synchronization (Timing and Frequency)
Let's first look at how synchronization is done in an isolated cell
environment first and then examine the issues under mobile/multi-cell
environments. Also, let's look at both the existing preamble design and
the alternative designs that we are aware of to the best of our knowledge.
In an isolated cell, it is well-known from literature how initial coarse
OFDM symbol timing and frequency offset can be estimated with the assist
of a repetition structure. Fine timing, such as through exploiting the
actual sequence rather than only the structure, can then kick in to
track the variations. Similarly, automatic frequency control (AFC) PLL
will then kick in for fine frequency tracking. A much less reliable
approach that I knew is to exploit the cyclic prefix to do coarse timing
and frequency offset estimation, but the method breaks down when the
channel is long (up to CP length for example) and it can only give
symbol timing info at its best, rather than frame synchronization that
can be achieved if the sync preamble structure is unique in a frame
(facilitate cell search).
I am not so clear on how initial timing and frequency synchronization is
done with the current preamble (100-500ms as stated by Yossi and Zion's
contribution for Runcom). The repetition structure of using every third
subcarrier does not hold if the FFT size is not a multiple of three.
Even say if we may compensate for it in the time domain samples so that
a repetition will appear, the structure is still lost when signals from
two segments are superposed. I guess timing synchronization is always
possible with a brute force sliding-window correlator (correlate with
all sequence candidates) over a frame duration if we can afford the
waiting time. But I am still fuzzy about frequency sync (pilot-based?
CP-based?) and whether we perform frequency sync before/after timing, or
simultaneously.
In a mobile/multi-cell environment, we are receiving from multiple
bases. The structure of the received signal is the same if all the
bases use the same structure (not necessarily the same sequence
though). The result is that the mobile will get synchronized with the
strongest base, which is fine during the first-time power-up. But the
timing should track the designated base, not always the strongest base,
so timing tracking should not rely on the structure (should utilize the
cell-specific sequences instead). Frequency tracking can use the
structure as well, but as said previously, it can lock to the strongest
base during handover, which can be a problem. In (quasi-)synchoronized
deployment, bases are locked to GPS clock in both time and frequency, so
locking to the strongest base in frequency may not be a big concern
(timing from bases can still be different because of propagation delay,
so lock to a weaker base can still be wanted). In the probably rare
case of driving directly away from one base and towards another, the AFC
can confuse Doppler with frequency offset, which may initiate frequency
re-acquisition. With the current preamble design, it may possible to
perform timing tracking in a similar fashion (i.e., using coherent
tracking schemes). I am not so clear on how frequency tracking can be done.
In summary, some kind of structure to facilitate initial coarse timing
and frequency synchronization is helpful. If the structure is only
helpful for limited scenarios with infrequent occurrence, the overhead
is something we have to consider. Maybe we should look for ways to
reduce the overhead. The simplest example is to reduce it to half a
symbol on average if we transmit such a signal every other frame. The
structured signal occupies the same subcarriers for all segments, so it
can be added in front of a preamble that uses orthogonal subcarriers for
three segments if PUSC is used in FCH and MAP transmission (but if FUSC
is used, we want interfering preambles between segments). The timing
tracking to one or more bases is better done through coherent schemes.
In order to estimate the channel characteristics to cells that occupy
the same set of subcarriers even though they can be one tier apart (for
soft handover or advanced interference suppression receiving), a low
correlation between preambles are desirable provided that we have
enough CP protection (micro-cell can be a good example for that capability).
2) MIMO channel estimation
If the default transmission for FCH and MAP is single-antenna
transmission, any preamble design with multi-antenna transmission will
degrade the estimation performance of the channel associated with that
default antenna. From this aspect, I think single antenna transmission
is fine. On the other hand, I do not think the current scattered pilots
are enough for MIMO channel estimation . So, it makes sense to me to
propose a MIMO-zone related single mid-amble symbol to obtain good MIMO
channel estimation. I am not sure if a mid-amble discussion is out of
the scope of this ad hoc group.
3) Operation mode identification
It makes sense to me if we tie a mode (PUSC, etc) with each FFT size,
which simplifies the "adaptive" mode identification problem. Could any
one help me to guess the default mode for 128-point FFT when we have
fewer subcarriers in one symbol than what a FCH needs? Maybe using more
than two symbols for FCH and MAP? If FUSC, the preamble should be sent
with a FUSC fashion.
4) Cell ID identification
As the email exchange between Jason, JY, and Yossi has indicated, the
questions of how much cell ID information we need to extract from the
preamble and how many cell ID we need to search may still need to be
clarified. I benefited from the discussion. I have an additional
question that is when sometimes we have two preambles defined by the
same cellID and segment (e.g., 0 and 0), which one do we use?
Best Regards,
- Jeff Zhuang
Motorola