FHSS Radio Design

T. C. McDermott, N5EGSpread Spectrum Seminar

October 12, 1997

Outline

• Design Objectives• System Level Requirements• Implementation Options• Implementation Details• Status

Design Objectives

• Provide ~ ISDN throughput (128 kb/s)• with options for greater throughput

• Path length of 20 miles• For Internet Protocol (IP) frames• Minimal disruption of existing users

• 902-926 MHz

• Accommodate reasonable number of simultaneous users

• Provide point-to-point and hub functions• Ethernet Interface to Computer• Future - Router Functionality Supporting TCP/IP

System Level

• Time-Division Half-Duplex (TDHD) is cost effective• No duplexers needed• Single antenna, feedline

• Radio to transmit for 10 milliseconds, then receive for 10 milliseconds• Minimize latency with fast switching

• Normally, time slots alternate between transmit and receive.

• Possible to have more transmit slots than receive slots when traffic is asymmetrical• Potentially doubles throughput

Throughput

• Modulation format: QPSK• Symbol rate: 300 ksym/s• Raw bit rate = 2 * 300k = 600 kb/s

10 milliseconds

Tx

RxMode 0

Tx

RxMode 1T

1 2 3 4 5 6 7 8 9 10 11 0 1

Tx

RxMode 1R

20 3 4

Tx

RxMode 2T

Tx

RxMode 2R

FEC Rate = 1/2

FEC Rate = 1/2

FEC Rate = 7/8

Mode 0, 1, 2: new hop each 10 msec slot

Throughput Calculations

skbskbMode

skbskbMode

skbskbMode

/424.94375 10

83.8

12

11

8

7/600:2

/825.24210

83.8

12

11

2

1/600:1

/45.13210

83.8

2

1

2

1/600:0

=•••

=•••

=•••

FEC Rate

Transmit Density

Overhead Factor

Hubbing Configuration

• Multiple radio configuration is useful (hub)• Concentrate users onto a single Internet connection• Provide link to a remote Internet connection

• Requires Synchronization between T/R switching of radios• All radios transmit, receive at same time (Mode 0)• One radio is in control

FHSSRadio

FHSSRadio

PassivePower

Combiner /Splitter

10-base-THub

To Internet

Highlinearity

PA

Filter + RX Amp

Hub Functions

• Multiple access (one radio is control channel, assigns users to data radios)

• Remote linking (occupy one data radio as link)• Spreading sequence the same for all radios,

but each starts at a different sequence offset

Radio

Radio

Radio

Radio

Control Chan

User Chan

User Chan

Link Chan Radio To Internet

Radio User

(Link)

(User)

System Gain Requirements

AssumptionsTx Output Power = 1.0 wattTx Antenna gain = 6 dB, Cable loss = 3 dBRx Antenna Temperature = 293KRx Antenna gain = 8 dB, cable loss = 3 dBFrequency = 915 MHz, Rx BW = 600 KHzRx NF = 8.0 dBCalculations • Tx ERP = 2.0 watts• Rx Noise Temperature = 438.4K• S/N at Rx = +21.1 dB.

(Eb/No = +18.1 dB for QPSK)

Eb/No Requirement

1E-12

1E-11

1E-10

1E-09

1E-08

1E-07

1E-06

1E-05

0.0001

0.001

0.01

0.1

1

0 0. 1 1. 2 2. 3 3. 4 4. 5 5. 6 6. 7 7. 8 8. 9 9. 10 10 11 11 12 12 13 13 14 14 15 15 16

Signal to Noise ratio, per bit ,

ate

Coherent 2PSK,

4PSK

Coherent 2FSK

For BPSK or QPSK, a BER of 10-6 requires +10.5 dB Eb/No

BER of 10 -9 requires +12.5 dB Eb/No

Differential QPSK degrades 2.3 dB, thus a BER of 10 -6 requires +12.8 dB Eb/No

Implementation Margin

Available Eb/No = +18.1 dBNeeded Eb/No at 10 -6 is +12.8 dB• Implementation Margin = 5.3 dB.Can improve Eb/No requirement by using

Forward Error Correction (FEC).Convolutional Codes (3-bit soft):Rate 1/2, 10-6 BER, needed Eb/No = 5.1 + 2.3 dBRate 7/8, 10-6 BER, needed Eb/No ~ 6.7 + 2.3 dB• Implementation Margin = 10.7 dB (rate 1/2)• Implementation Margin = 9.1 dB (rate 7/8)

Feedline Loss at 900 MHz

• RG58 = 16 dB / 100 feet• RG8 = 6.7 dB / 100 feet• 9913 = 4.2 dB / 100 feet

IF & RF Diagram

ComplexUp / DownConverterRx Limiter

I

Q

I

Q

LocalOscillator

85.35 MHzSAW Filter

HoppingVCO 1

HoppingVCO 2

915 MHzRF Filter

LNA

PA

915 MHzRF Filter

T/RSwitch

Sin+Cos

Baseband Diagram

MotorolaMC68360processor

MotorolaMC68160Ethernet

10-base- TLAN I/F

QualcommConvolutionalFEC

FLASH+ DRAM

BasebandAnalogInterface

DigitalQPSKDemod

QPSKModulator& Filter

A-to-DConverter

I

Q

A-to-DConverter

I

Q

Receiver Front-End Calculations

Stage 1 2 3 4

Filter+T/R LNA+Mix IF filter IF Post Amp

Noise Figure (dB) 0 5.5 0 8

Gain (dB) -3 28 -15 18

Pout (dBm) -111 -83 -98 -80

Input Power (dBm) -108

Cascade NF (dB) 7.84

Temperature (C) 25

BW (kHz) 650

Noise Power (dBm) -108

Noise voltage (uV) 0.9 ( @ 50 ohms)

BPFT/R Sw LNA IF SAWFilter

Amp

915±13 MHz 600 KHz

1 2 3 4

85.35 MHz IF

Frequency-Hopping VCOs

• Frequency Hopping is achieved by altering the Local Oscillator frequency. Otherwise, the radio looks like a non-SS radio.

• Key issue is ‘settling time’ of the VCO’s.• This radio changes frequency each 10

milliseconds.• If there is one VCO, it must settle much, much

faster than 10 milliseconds -- this is very difficult.

• Solution: use 2 VCO’s. One slews to a new frequency whilst the other is being utilized. This allows 10 milliseconds settling time.

• 3 VCO’s would allow 20 msec settling time, etc.

Data Rate vs. SAW bandwidth

2x Symbol Rate

Transmit Spectrum - No filtering

Transmit Spectrum - alpha = 1 filtering

2x Symbol Rate

Transmit Spectrum - alpha < 1 filtering

(1+α) x Symbol Rate

α = 1.0, 300 ksym/s=> 600 kHz

α = 0.4, 300 ksym/s=> 420 kHz

Low-alpha raised-cosine filters narrow the emission bandwidth, minimize interference

from adjacent channels

FHSS - key issues

• Top three: • Speed• Speed• Speed

• VCO lockup time⇒ Multiple VCOs

• Carrier Recovery Lockup Time⇒ Memorize (compute) frequency error [make a really

good guess]

• Symbol Recovery Lockup Time⇒ Usually faster than carrier recovery lockup time

FEC / QPSK Modulator

• FEC doubles the bit rate (for rate = 1/2). Input is 300 kb/s, output is 600 kb/s.

• Each ‘pair’ of bits are used to select one of four phase states, producing 300 ksym/s.

• QPSK modulator filters the baseband signals, and differentially-encodes them. This coding means that the phase output difference is proportional to the symbol value.

ConvolutionalCoder

300 kb/s

600 kb/s

DiffEnc

FIR Filt

FIR Filt

300 kb/s I

300 kb/s Q

QPSK Modulator Fwd Err Correctionα = 0.4 rolloff

D-A Conv

D-A Conv

QPSK Upconversion

I

Q

Oscillator0-degrees

90-degrees

+ IF-output

I

QConstellation point (1,1)

1

0

1

0

QPSK Demodulator HSP50210

ComplexMult

NCO

CosSin

I

Q

RRC Filter

RRC Filter

I

Q

I

Q22 QI +

I

Q1tan−

Mag

Phase

Phase Error Det

Freq Error

Freq Sweeper

Loop Filter d/dt

AGC Loop

Symbol Error Det Sym Err

Slicer

Slicer

Isoft

Qsoft

Clock Recovery

ExpectedSymbol Values

TransitionMid-point

Sampling Error

Sampling too early (positive slope)

TransitionMid-point

Sampling Error

Sampling too early (negative slope)

ExpectedSymbol Values

Sign of error term depends on direction of slopeas well as early / late timing of the sample

Mid-symbol samples

End-symbol sample

Clock Recovery Process

HSP 50210Demodulator

A-D Conv

A-D Conv

Sampling (Position) Error

Reg

iste

r

+Sign Bit

+

Con

stan

t

I

Q

Ana

log

Inpu

t

Numeric-Controlled Oscillator (NCO)

Loop Filter

Carrier Recovery

I

QExpected constellation point

Actual constellationpoint

θ error

I

Q

θ error

Rotation by 45 degreesInitial Constellation

I

Q

θ error

Multiplication by 4

1. Rotation,2. Multiplication by 4.This Removes modulationcomponents from the error term,and references error to thezero-phase position.

Carrier Recovery

• Carrier Error Detector implements ‘Frequency Error’ term (how much the phase rotates each sample).

• If the loop is too far off-frequency, the phase error accumulation increases too fast and the loop cannot lock.

• Sweep-aided acquisition allows the carrier reference oscillator to ‘sweep’ up and down in frequency. When the frequency gets close enough, the sweeper disconnects, and the loop acquires by phase alone.

• Sweeping takes a long time, and is to be avoided (except maybe during link setup).

s-plane vs. z-plane

• Frequency analysis on s-plane is conducted over j ω (the imaginary axis).

• Left-hand plane of s-plane maps to unit-disk of z-plane

• Thus, frequency analysis on z-plane is conducted over unit circle

s = jω z = ejωtRegion of

stability of poles in s-plane

Region of stability of poles

in z-plane

Loop Filter Analysis

• Frequency & Phase Response of digital filters most easily analyzed in Z-domain

• Apply Z-transform by inspection (a register is equivalent to multiplying by z -1)

• Derive Amplitude & Phase by evaluating H(z) on the unit circle in the Z-plane [substituting z = ejωt and varying ω from zero to 2 π/t]

• Excel spreadsheet can evaluate and plot results.

z = ejωt

Lead Lag Filter

Reg

iste

r

+

+

Filter Clock

In Out

Lead Gain

Lag Gain

Digital Multiplier

Filter Analysis

Klead

+ +Klag z-1W V

X

Y

Y = Klead X + V

V = Klag W z-1

W = X + V

V = Klag (X + V) z-1 = Klag X z-1 + Klag V z-1

V ( 1 - Klag z-1 ) = X Klag z-1

V = X Klag z-1

1 - Klag z-1

Y = X Klead + X Klag z-1

1 - Klag z-1

YX

=Klag + Klead ( 1 - Klag) z-1

1 - Klag z-1H(z) =

Loop Filter PerformanceKlead = 20

Klag = 0.98

-10

-5

0

5

10

15

20

25

30

35

40

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Normalized Frequency, n/T

-90

-45

0

45

90

mag H(s), dB

phase H(s),

degrees

NCO - Phase response

• NC-VCO converts error signal to accumulating phase (i.e.: frequency).

• Need to derive relationship of output phase to input error signal.

Register (z

-1)

+In Out

( )

( )

1

1

1

1

−=

=−+=

+= −

zIn

Out

InOutz

OutInOutz

zOutInOut

(thus: phase = integral of frequency vs. time)

Carrier Loop

• Loop Filter resides inside Carrier Recovery Loop. Closed-loop transfer function needs to be computed.

Loop FilterBasebanddata

NCO

Phase ErrorCalculation

RecoveredCarrier

H(z)ϕe

ϕe filt

Kp

( )( )

NH

NH

NHNH

HN

i

o

io

oio

+=

=+−=

1

1

ϕϕ

ϕϕϕϕϕ

)(1

)(1

zHKz

zHKz

KN

p

p

i

o

p

+−=

−=

ϕϕ

A poor choice for filter constants

-25

-20

-15

-10

-5

0

5

10

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Frequency

dB

.

-180

-135

-90

-45

0

45

90

135

180

Loop Response - Track Mode

-80

-70

-60

-50

-40

-30

-20

-10

0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Frequency

dB

.

-180

-135

-90

-45

0

45

90

135

180

Track Mode - High Resolution

-6

-5

-4

-3

-2

-1

0

0.00E+00 1.00E-04 2.00E-04 3.00E-04 4.00E-04 5.00E-04 6.00E-04 7.00E-04 8.00E-04 9.00E-04 1.00E-03

Frequency

dB

.

-70

-60

-50

-40

-30

-20

-10

0

Loop Response - Acquisition Mode

-80

-70

-60

-50

-40

-30

-20

-10

0

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Frequency

dB

.

-180

-135

-90

-45

0

45

90

135

180

Overall T/R timing

Pd

2Pd

Pd

Txp

Txp

Start Tx

Stop Tx

Start Tx

Stop Tx

( )dp

pd

PTcycleTx

TxPcycleRT

−=

+=+ 22

For Pd = 110 µsec (20 miles):Txp = 10 msec - 110 µsec = 9890 µsec transmit duration

Receiver recovery time = 2 P d = 220 µsec (+ tx ramp time)

Hub End User End

Receiver Acquisition Timing

Receive

Hub End

User End

Transmit

100 µsec transmitter power rampup800 µsec carrier recovery lockup< 800 µsec symbol recovery lockup

60 µsec Viterbi decoder lockup

960 µsec total receiver lockup time

Detailed Timing

Hub End

User End

100 µsec transmitter power rampdown

110 µsec:

210 to 320 µsec receiver recovery(prop delay + Tx rampup)

ReferenceEpoch

960 µsec receiver lockup

10000 µsec 10000 µsec

9890 µsecTx duation

8830 µsecdata portion

20 miles: propagation delay0 miles: channel idle time10 miles: half prop delay, half idle time

Receive

Transmit Receive

Transmit

10000 µsec 10000 µsec

T/R timing

• Hub establishes 10.000 millisecond period from it’s internal clock.

• Timing epoch is the opening FLAG from the Hub radio transmit frame, 960 µsec after the transmission starts.

• User radio establishes timing by listening for the hub epoch, smooths arrival times.

• Transmitter ramp-up and ramp down consume 100 milliseconds each.

• Transmission duration is 9,890 µsec, data portion is 8830 µsec.

• Propagation delay is measured during link startup. Hub computes the delay, communicates to user.