diff options
Diffstat (limited to 'sfn.tex')
-rw-r--r-- | sfn.tex | 149 |
1 files changed, 149 insertions, 0 deletions
@@ -0,0 +1,149 @@ +\section{Single-Frequency Networks} +\subsection{Requirements} +The DAB standard has been designed to enable the creation of transmission +networks where several transmitters share the same frequency, and send the same +signal synchronously. Such networks are called ``Single-Frequency Networks''. +Each transmitter needs to be fed the same multiplex stream, which must include +timing information required for synchronisation. This timing information implies +that a time reference must be installed at each transmitter. + +The requirements for a SFN can therefore be summarised in three points: +\begin{itemize} + \item The signal must be \emph{identical} for each transmitter. This + requires a common multiplexers, and a distribution network that carries + the ETI to all modulators. + \item All transmitters must transmit on the \emph{same frequency}. The modulators + require a frequency reference. + \item The signal must be transmitted at the \emph{same time}, which requires + a time reference at each site. It also implies that the ETI stream must + contain timestamps. +\end{itemize} + + +The figure~\ref{fig:txchain-sfn} shows a SFN setup with two transmitters. + +\begin{figure}[h] + \includegraphics[width=\textwidth]{figures/txchain-sfn.pdf} + \caption{This outline for a SFN shows two transmission sites.} + \label{fig:txchain-sfn} +\end{figure} + +\sidenote{Explain requirements on system time, NTP} + +\subsection{Multiplexer Configuration} +On the ODR-DabMux configuration, there are not many options that are specific to +an SFN setup. +Most importantly, the timestamp feature must be enabled using the ``tist'' option in +the ``general'' section. + +Furthermore, it is recommended to use the ZeroMQ transport between the +multiplexer and the modulators, which can be enabled in the ``outputs'' section. +Care has to be taken to have an output that slows ODR-DabMux down to nominal +rate. The ZeroMQ output alone does not enforce this. The following listing shows +the relevant options we just covered. + +\begin{lstlisting} +general { + tist true + ... +} + +... + +outputs { + ; Accept connections on all interfaces, on port 9100 + zmq "zmq+tcp://*:9100" + ; This throttles muxing down to nominal rate + throttle "simul://" +} +\end{lstlisting} + +\subsection{Modulator Configuration} +Since the modulator has to ensure that the three SFN requirements are satisfied, +its configuration is more complex. + +We will assume, in this explanation, that one of the following USRP devices is +used: USRP2, USRP B100, USRP B200. Other devices also support the necessary time +and frequency synchronisation, but they have not been well tested. These USRP +devices can accept different sources for the reference clock: +\begin{itemize} + \item The default ``internal'' source uses the non-disciplined + clock generator inside the USRP. It is not suitable for SFN. + \item The ``external'' source corresponds to the SMA connector on + the USRP. A 10MHz signal from an external source must be connected to + it. + \item The optional GPSDO that can be mounted inside the USRP, and is + selected as source with the ``gpsdo'' setting. +\end{itemize} + +For the time reference, the ``pps\_source'' option is used. Possible values are +``none'', ``external'' and ``gpsdo'', with analogous meaning as for the +reference clock. + +In case the USRP is connected to external references, the relevant configuration +would be as follows: + +\begin{lstlisting} +[uhdoutput] +refclk_source=external +pps_source=external +\end{lstlisting} + +These settings alone do not tell the modulator to enable synchronisation of the +transmission, they only select how the USRP is configured. To enable timestamp +decoding and the frame synchronisation logic in ODR-DabMod, the following +settings must also be set: + +\begin{lstlisting} +[delaymanagement] +synchronous=1 + +; The constant offset to be added to the TIST, in seconds +offset=2.0 +\end{lstlisting} + +The ``offset'' setting deserves some further explanations. The ETI data stream +contains TIST information, from which a time-stamp for each ETI frame can be +derived. Each ETI frame ($24$\ms interval) is therefore associated with a +precise point in time that defines the time of transmission of the corresponding +transmission frame.\footnote{It is slightly more complex, because one + transmission frame is composed of several ETI frames in some + transmission modes, but the principle stays the same. It suffices for this + explanation that we can derive the transmission time from the TIST +information.} The TIST information is set to current time at ETI frame +generation, and does not take in account the propagation delay across the +distribution network. Therefore, we need to add an offset, called $\delta$, to +the TIST to define transmission time. + +\[ +t_{transmission} = t_{TIST} + \delta +\] + +If this offset is set to a higher value, there will be a bigger delay (measured +in absolute time) between the point in time a frame is multiplexed and the point +in time the frame is transmitted. More frames therefore will be buffered in +the ODR-DabMod ZeroMQ input, increasing robustness against network latency +fluctuations. + +The offset already has two functions: it compensates for network delay and +allows a trade-off between delay and robustness. But it also serves a third +purpose: When doing coverage planning for an SFN, it is necessary to be able to +control the relative delay between transmitters in the order of milliseconds. +This tuning of relative delay is included in the ``offset'' setting. We can +therefore rewrite the above equation as: + +\[ + t_{transmission} = t_{TIST} + \delta_{network} + \delta_{planning} +\] +\[ + \delta = \delta_{network} + \delta_{planning} +\] + +\sidenote{Explain relationship with ZeroMQ max buffer size} + +\subsection{Using Ettus GPSDO} +\subsection{Using ODR LEA-M8F GPSDO board} + +\sidenote{Give example} + +% vim: spl=en spell tw=80 et |