Timed Synchronous Dataflow

*Telecommunications and Signal Processing Seminar*: Mr. Mark Grechanik, UT Austin, Friday, 3:00 PM, ENS 637. Food and drinks provided.

- Heterogeneous in the algorithms used (modulation/demodulation, encoding/decoding, fixed/adaptive filtering, etc.) and implementation technologies (digital VLSI, microcontrollers, digital signal processors, mixed signal IC, and RF circuits),
- Advanced Design System (ADS) by Agilent EEsof is a cohesive cosimulation framework for the more than 70 simulators by Agilent EEsof. The cohesion is made possible by the Timed Synchronous Dataflow (TSDF) model, which efficiently cosimulates mixed RF, analog, and digital hardware/software subsystems. Like SDF, TSDF can be statically scheduled: all analysis can be performed at compile time. Agilent EE sof ADS Web page

- A canonical representation of a communication signal is
s(t) = I(t) cos(2 pi f

where f_{c}t) - Q(t) sin(2 pi f_{c}t)_{c}is the carrier frequency, I(t) is the in-phase component, and Q(t) is the quadrature component. The notion of in-phase and quadrature components arise from the fact that the sine term is 90 degrees out of phase with the cosine term. - Common carrier frequencies:
- 800 MHz - 2.5 GHz for mobile terrestrial wireless communications
- 88-108 MHz for FM radio
- 550-1700 kHz for AM radio.

- The in-phase and quadrature components generally contain low frequencies,
e.g. 0-4 kHz for speech or 0-20 kHz for audio.
We use f
_{m}to denote the highest frequency in the baseband component, and f_{c}is assumed to be much greater than f_{m}. - The canonical representation is a time-frequency representation
composed of
- Time (baseband) information: I(t), Q(t), and t
- Frequency information: f
_{c}

_{m}/ (f_{c}+ f_{m}). For the case of IS-95 cell phones, the savings is 34000:1.

- Adds a notion of time to the SDF model
- Adds two new timed data types based on the canonical representation of
communication signals described above.
The baseband is sampled every delta t time units.
A communication signal is represented by the following data structure:
- Envelope type
- sampled I(t)
- sampled Q(t)
- delta t
- f
_{c}

- Baseband signal
- sampled I(t)
- sampled Q(t)
- delta t

- Envelope type
- Extra attributes
- timed arcs: delta t
- timed actors: fired at a constant rate
- input/output ports: optional impedance

- Specifying delay for a timed particle data type requires the definition
of a "0" time particle data value
- I = Q = 0
- delta t and f
_{c}are to be determined by the scheduler

- Scheduling [2]
- load balancing to find the repetitions vector: linear algorithm
- check sampling intervals for consistency: linear algorithm
- check carrier frequencies for consistency (heuristic):
O(
*n*^{3}) algorithm to find the topological sort, and then analyze the graph returned by the topological sort either once or two - schedule the graph: worst case is exponential of a polynomial in
*n*, but PGAN/RPMC heuristics for generating a uniprocessor schedule is O(*n*^{3})

- Multicarrier modulation
- As a timed particle, choose the midpoint of the carrier frequencies and remodulate the envelopes
- In circuit envelope simulation, the simulator handles multiple carrier frequencies at once.

- Jose Luis Pino, Personal Correspondence, March 31, 1999.
- Jose Luis Pino and Kal Kalbasi,
"Cosimulating Dataflow with Analog RF Circuits",
*Proc. IEEE Asilomar Conference on Signals, Systems, and Computers*, Pacific Grove, CA, Nov. 1998.

Updated 02/18/02.