1 / 11

Thermodynamics of A/D Conversion

Thermodynamics of A/D Conversion. Problem. What is the minimum power required to perform n-bit A/D conversion on a signal with a certain bandwidth? Would like a bound independent of technology and A/D architecture But the flip side is that THIS BOUND MAY NOT BE PRACTICAL

borka
Download Presentation

Thermodynamics of A/D Conversion

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Thermodynamics of A/D Conversion

  2. Problem • What is the minimum power required to perform n-bit A/D conversion on a signal with a certain bandwidth? • Would like a bound independent of technology and A/D architecture • But the flip side is that THIS BOUND MAY NOT BE PRACTICAL • Only would like to know what’s physically possible and what’s not • Eventually need to consider power required to overcome: • Thermal noise • Mismatch • Clock jitter • and power required for discretize the analog voltage

  3. Thermal Noise • Most obviously causes power dissipation in the sample and hold circuits • Have to spend energy to overcome <Vn2> = kT/C noise • If the stream of signal samples is uncorrelated and has variance of <Vin2>= SNR·<Vn2> … • …we spend average energy <E> = C·<Vin2> = kT·SNR per sample • But is kT/C noise unavoidable? • Even if it is, can we construct a more efficient sampler?

  4. kT/C Noise • Capacitance is not the only memory element • Can use capacitance, inductance, mass on a spring – anything that can store energy • But as long as there is dissipation (equivalent of a resistor) in the system, there is still equivalent noise! • Thus <E> = kT·SNR per sample still has to dissipated • Note that that it’s independent of C, L, kspring, etc. • But does any sampler have to have a dissipative element (resistor or equivalent)?

  5. Dissipation in a Sampler • For example, try to construct a capacitive sampler without a resistor: • Cannot transmit any meaningful information to the capacitor, because system has infinitely narrow bandwidth • Can charge and discharge the capacitor without any dissipation • But can not erase the previous sample without memorizing what it was somewhere else • Fundamentally need a “trash can” to dump memorized information • This follows from the 2nd Law of Thermodynamics

  6. “Trash Can” • In electrical circuits, “trash can” is usually a thermal environment • We dump out old information by connecting to a dissipative element (resistor) • But then we always pick up thermal noise • Therefore, ANY system that memorizes input information must have thermal noise and dissipate energy to overcome this noise • Can we construct something more efficient than normal sampler? • Yes, by making it reversible • means at every step process can go either forward or back • then no energy is needlessly wasted and maximum efficiency is obtained • similar to Carnot cycle in thermodynamics

  7. Carnot Sampler • Step 1 – connect C=C1 to voltage source and slowly ramp the voltage from 0V to Vsignal: • Can deposit signal charge qsignal=C1·Vsignal onto C with arbitrary low energy loss in R by making ramp rate >> RC1 • Voltage source only supplies average energy W1= <qsignal2> / ( 2C1 ) • Step 2 – disconnect C from the input and reset voltage source to 0V: • Signal charge is memorized on the capacitor together with thermal noise <qnoise2> • Average energy stored on C=C1 is • <E1>= (<qsignal2>+ <qnoise2>) / (2C1) • SNR = <qsignal2> / <qnoise2>

  8. Carnot Sampler • Step 3 – let C increase from C1 to C2 and use the energy released in this process: • Charge remains the same, but average energy stored reduces to • <E2>= (<qsignal2>+ <qnoise2>) / (2C2) • Want to make <E2> equal to energy of the thermal environment kT/2 • The excess energy W2=<E1>-<E2> can be recovered as useful work • Step 4 – connect C=C2 back to the signal source: • Average energy <E2> stored on the capacitor is equal to the average energy of thermal noise kT/2 • Thus, no energy exchange takes place, and no energy is wasted

  9. Carnot Sampler • Step 5 – prepare for the next sample by decreasing C from C2 back to C1: • Requires external work • W3 = 0.5 · kT · ln( C2 / C1 ) • This work is dissipated in R and cannot be recovered • The total average energy spent per sample is only • Qwaste = W1-W2+W3 = 0.5 · kT · ln( C2 / C1 ) • But C1 = <qnoise2> / kT, and C2 = (<qsignal2>+<qnoise2>) / kT • Thus, Qwaste=0.5 · kT · ln((<qsignal2>+<qnoise2>)/<qnoise2> )= • Qwaste=0.5 · kT · ln( SNR + 1 ) is the absolute lower bound on sampler energy dissipation per sample

  10. Quantization • So far we only considered analog sampling • Similarly, some energy fundamentally has to be spent for memorizing and erasing digital outputs as a function of bit error rate • But do we really have to “memorize” analog value? • Do we really have to “memorize” digital outputs? • Does energy fundamentally have to be spent on any other intermediate process? • Not sure, but have some ideas…

  11. Conclusions • Found theoretical lower bound on energy required to store and erase analog signal with certain SNR • Still need to figure out what’s the equivalent bound on energy for quantization • What are the other possible fundamental sources of energy dissipation? Jitter? Mismatch?

More Related