class notes

5539:
Switch: is black box
Radio interface:(Dimensioning) (limiting step)
Datapath between client and mobile station
vs
Backhaul:
Datapath between different base mobile stations: fixed line(backhaul)

Radio interface is limiting step because once a ch

Information is represented by information entropy

Signals: Periodic signals(frequency/angular frequency/cycle, amplitude & phase)
Asin(ωt+θ)=[A exp(jθ) exp(jωt)]/[Z exp(jωt)](Z is called complex amplitude)
Sum of 2 periodic signals is not periodic if it involves irrational numbers.
Aperiodic signals: unit step, unit impulse, unit ramp

ADC: Discretizaion of time(sampling), or s(t)(quantisation)
Sampling: Nyquist sampling theorem: fs > 2W(highest frequency). If fs < 2W aliasing occurs.
Quantisation is irreversible.
quantisation: N bits, 2^N levels.
E.g. 44.1khz, 16 bits, 2 stereo channels, 1.411MB per second, so we use compression algorithms such as mp3 or flac.

Analog: Noise directly affects signal quality, SNR is increased by amplification.
Digital: SNR is now irrelevant to communication(binary now), amplification is no longer necessary.

5000:
More transistors give more performance, but are more difficult to design, and more expensive to design and manufacture.(E.g. server processor are more expensive)

Design: area of chip,

transistor is turned by controlling gate voltage(field effect). resistance is essential to controlling to channel, so polycide is used to control it. In strained silicon, other metals is added distance to silicon lattice to increase conductivity. Originally we used metal gates(MOS) that are faster, but now we use poly silicon gates(POS) because they are made faster. but we switched back to MOS because POS had a lot of resistance and didn't have good performance. Smaller transistor area, smaller resistance, larger current with same voltage, faster switching speed(e.g. clock rate), also more transistors from same wafer, less cost. metal is insulated from silicon, and it's a switch. between metal and silicon

5650:
computers are cheap so digital wins over analog

5060:
voltage controller is used for power supply

bus is used for communication trasnfer

cross bar structure: costs increases exponentially as components amount increase(n!/2 connections for larger systems)

5000:
carrier recombination: 1 electron filling a hole, and leaving another hole in silicon doping.

in NMOS, every gate is connect to lowest reference voltage on the chip(a.k.a. GND); in PMOS it's highest.

smaller transistors, higher speed/current, more parallelism, but more leakage, more power consumption
wider transistors consume more power

5538:
Rectangular pulses consume more power than wave pulses, so we use a pulse shaping filter for every signal. Bit rate is the same though.

Energy=0.5*signal 1 Energy(integrate signal energy over time)+0.5*signal 2 Energy(integrate signal energy over time)
Power=Energy/Time
Increase bandwidth by decreasing symbol time(decreasing bandwidth too), or using a larger symbol set(E.g. Quaternary modulator takes in 2 bits to create 4 possible waveforms, bandwidth stays the same, but higher bit error rate)
bit rate(bits per second)=bits per symbol*bandwidth(symbols per second)
baud rate(symbol rate)=1/symbol duration

5060:
2 types of sensors: chemical and optical

xnor.ai: ai2go compresses machine learning algorithms to smaller scale so they can work on other hand-held embedded devices and makes ai more deployable.

cache memory can be accessed in the same cycle.

5000:
IDS_NMOS
ISD_PMOS(inverted)

5539:
Error rate down, judgement overlap area down, dmin(minimum distance btw 2 points) up
bpsk & bask: same bandwidth and bit rate
bpsk: half(3db) of energy for same bit error as bask.
bpsk has better bit error rate(collinear, same magnitude, separation is 2), ask has lesser power consumption(one of the signals has zero energy and zero norm)

question: bpsk has half energy but same dmin as bask, conserve more energy?

bfsk also is equidistant, they are orthogonal so separation is not large as bpsk(separation is sqrt 2)

bpsk has largest dmin compared to bfsk or bask for same energy.

inner product of two vectors positive: angle is acute

qpsk is also called iq modulator

X-qam is better than X-psk because dmin is larger

higher qam gives you more bandwidth but more ber, so it's used where SNR is large(E.g. Ethernet & cable TV), not when SNR is low(E.g. cellular networks and Wi-Fi rarely have high dimensions)

high bandwidth efficiency: mqam

high energy efficiency: mfsk(double frequencies require double bandwidth 16PSK> 16 QAM, Same Bandwidth and ber, 16 psk and 16fsk not so straightforward)

---------IMPORTANT! EXAM QUESTIONS WILL ASK YOU TO MAKE DESIGN DECISIONS BASED ON CONDITIONS! SUCH AS DETERMINING MQAM VS MFSK IN SATELLITES(mfsk because of energy transportaion costs) AND TERRESTRIAL STATIONS(mqam because of bandwidth licensing costs)!---------

5000:
current increase: decrease sd distance or increase sd voltage
velocity saturation: velocity of electrons increase up to a saturation point, which cause current to saturate.
nmos pass 0, cmos pass 1

5539:
Power is O(d^(-2))(order of, or inversely proportional)


Path loss dynamic range: 2km ~ 140 dB!!(speed of signal amplitude change due to distance change is slowest)
Shadowing dynamic range: 8-10dB(slow)
Multipath: waves from different path causes con/de-structive inteference(similar to double slit experiment) dynamic range: 40dB according to Rayleigh distribution (fast)
We are working with Linear Time VARIANT System in fading.

5060:
every hardware componene that has firmware has a processor, e.g. CPU talks to intermediary processor in hdd to store data in hdd

memory is often structured in square blocks of arrays because otherwise chips would be too long

5000:
change body doping to change threshold voltage

presence of foreign wells changes threshold voltage in a direction that decreases the voltage

electron mobility technology are more developed than that of hole mobility, so NMOS is preferred(this class assumes n well by default)

select/active: select is concentrated area of doping used for transistor contacts, active is less concentrated area of doping used for wells

yield=no. of good chips/no. of total chips

chips are not getting smallers because error of layout is bigger, so channel and diffusion length error, and then current and transistor speed error is bigger

5000:
m1-5 are used for different purposes so they are different(m1&2 is often used for short distance)

5650:
multipath effect is frequency selective

ISI is caused by echo(P of matrix is related to delay spread)

y5=h0s5+h1s4+h2s3+h3s2+n5(p=4)

5539:
selection combining has 3db penalty

receiver composed of RF, ADC, & most of all, baseband DSP.

each diversity signal requires 1 RF & ADC that is combined into 1 MRC, so we usually use only 1 RF ACD & baseband with a switch between multiple signals. We use a preamble to estimate/switch between the antenna(preamble is usually time divided). More antennas, so estimation/switching between antenna is less accurate.

5000:
weak 0 & weak 1 is caused by vdd-vtn and vgnd+vtp: threshold voltage

Vgs has to be large enough

larger diffusion area, waste silicon, take up too much space, more electronc travel distance & resistance, lesser electron speed & current

5000:
load capacitance of nand w/ 3 inverters is different than nand w/ 1 inverter so speed is different

given same shared diffusion area, it's better to have as little contacts as possibe because contacts take up a lot of area.

5000:
complementary networks are not used sometimes because or/and networks can take up more metal, need more gates than transmission gates, or other reasons.



5000:
Velocity saturation causes resistance to be constant as voltage increases
PMOS res>NMOS res because electron mobility>hole mobility
NMOS velocity saturation reached w/ smaller voltage

Critical path delay: delay of longest path in CMOS system(limiting steo)
Cpacaitors are placed in switching nodes btw transistors
Different place use different gates depending on requirement of gate's place

parasitic capacitance caused little voltage overshoots because you need time to discharge capacitance

 ceramic capacitors can't be used in IC because too large, we use MOSFET or MIM(more accurate) for capacitance.