Matlab Model Rf

RF Component System Objects (Complex Baseband)

Process complex baseband RF signals through standalone System Objects -- amplifiers, mixers, filters, S-parameter models, and power amplifiers with memory. All objects follow the standard System Object lifecycle (step/reset/release) and process data entirely in MATLAB with no Simulink dependency.

Requires R2024b+ (rf.PAmemory: R2024a+).

System Object Lifecycle

All rf.* System Objects share this pattern:

obj = rf.Amplifier(Gain=15, OIP3=35);   % Create
out = obj(in);                            % Step (process one frame)
reset(obj);                               % Reset internal state
out2 = obj(in2);                          % Step again
release(obj);                             % Release (allows property changes)
obj.Gain = 20;                            % Modify properties
out3 = obj(in3);                          % Re-setup on next step

Properties are locked between first step and release. Attempting to change a Nontunable property without calling release first errors. Tunable properties (like Gain) can be changed between step calls.

rf.Amplifier -- Nonlinear Amplifier Models

Construction

amp = rf.Amplifier;                                    % All defaults
amp = rf.Amplifier(Gain=15, OIP3=35);                 % Polynomial (default)
amp = rf.Amplifier(Model='ampm', Table=ampmData);     % AM/PM table
amp = rf.Amplifier(Model='modified-rapp', MagnitudeGainDB=15, Vsat=1.5);
amp = rf.Amplifier(Model='saleh', InputScaleDB=-10);

Nonlinear Models

Matching P1dB across models: Only poly/cubic accept OP1dB directly. For modified-rapp and saleh, find P1dB by sweeping input power. See reference/amplifier-nonlinear-models.md for the numeric sweep recipe, AM/AM comparison across model types, poly-vs-cubic details, compression I/O conventions, and Nonlinearity property gotchas.

Noise

amp = rf.Amplifier(Gain=15, OIP3=35, ...
    IncludeNoise=true, NoiseType='NF', NF=2, ...
    SampleRate=10e6);

Visualization

amp = rf.Amplifier(Gain=15, OIP3=35);
visualize(amp);   % Plots AM/AM and AM/PM curves

Example: 16-QAM Through Nonlinear Amplifier

amp = rf.Amplifier(Gain=15, OIP3=35);

% Generate 16-QAM signal
rng(42);
data = randi([0 15], 1000, 1);
modSig = qammod(data, 16, UnitAveragePower=true);

% Scale to desired input power level (1-ohm normalized: Vin = sqrt(Pin_W))
Pin_dBm = -20;
Pin_W = 10^((Pin_dBm - 30) / 10);
Vin = sqrt(Pin_W);
inSig = Vin * modSig;

% Amplify
outSig = amp(inSig);

% Measure (1-ohm: Pout = |Vout|^2)
Pout = 10*log10(mean(abs(outSig).^2)) + 30;
gain = 10*log10(mean(abs(outSig).^2) / mean(abs(inSig).^2));
fprintf('Gain: %.2f dB, Pout: %.2f dBm\n', gain, Pout);

release(amp);

rf.Mixer -- Modulator/Demodulator with Impairments

Construction

mix = rf.Mixer;                                       % All defaults (modulator)
mix = rf.Mixer(Model='demod', Gain=-6, RF=2.4e9, LO=2.1e9);
mix = rf.Mixer(Model='iqmod', Gain=-8);               % No RF or PhaseOffset needed
mix = rf.Mixer(Model='iqdemod', Gain=-6, RF=2.4e9);  % RF relevant; no LO or PhaseOffset

Models

Properties

Impairments

mix = rf.Mixer(Model='iqdemod', Gain=-6, RF=2.4e9, ...
    GainImbalance=0.5, ...     % I/Q gain imbalance (dB)
    PhaseImbalance=2);         % I/Q phase imbalance (degrees)
% Note: LO and PhaseOffset are not relevant for iqmod/iqdemod -- omit them

Nonlinearity

mix = rf.Mixer(Model='demod', Gain=-6, RF=2.4e9, LO=2.1e9, ...
    Nonlinearity='IIP3', IIP3=10);

Same compression options as rf.Amplifier: IIP3, OIP3, IP1dB, OP1dB, IPsat, OPsat.

Phase Noise

mix = rf.Mixer(Model='demod', Gain=-6, RF=2.4e9, LO=2.1e9, ...
    IncludePhaseNoise=true, ...
    PhaseNoiseLevel=[-80 -100 -120], ...           % dBc/Hz at offsets
    PhaseNoiseFrequencyOffset=[1e3 10e3 100e3], ... % Hz offsets
    SampleRate=10e6);   % SampleRate only relevant when noise/phase-noise enabled

Visualization

visualizePower(mix);       % Plot power characteristics
visualizePhaseNoise(mix);  % Plot phase noise spectrum

Gotcha: For 'iqmod', RF and PhaseOffset are irrelevant (emit "not relevant" warnings). LO does not warn and does not affect frequency conversion, but it scales phase noise when IncludePhaseNoise=true -- set it to the intended carrier frequency only if modeling phase noise. For 'iqdemod', RF IS relevant (no warning), but LO and PhaseOffset are not relevant and emit warnings. This asymmetry is because iqdemod needs the carrier reference for demodulation while iqmod operates purely at baseband (no frequency conversion).

Example: IQ Demodulation

mix = rf.Mixer(Model='iqdemod', Gain=-6, RF=2.4e9, ...
    GainImbalance=0.3, PhaseImbalance=1.5);

% Simulate received RF signal (complex)
t = (0:999)' / 10e6;
rfSig = exp(1j * 2*pi * 5e6 * t);   % 5 MHz tone at carrier

bbOut = mix(rfSig);                   % Complex baseband output with impairments
release(mix);

rf.Filter -- Complex Baseband Filter

Construction

filt = rf.Filter;
filt = rf.Filter(DesignMethod='Butterworth', ResponseType='Bandpass', ...
    PassFreq_bp=[2.395e9 2.405e9], RF=2.4e9, SampleRate=100e6);

Frequency Convention (Critical)

All PassFreq_* and StopFreq_* properties specify absolute RF frequencies (not baseband offsets). The RF property defines the carrier center, and the filter internally translates to baseband representation.

% CORRECT: 10 MHz channel filter centered at 2.4 GHz
filt = rf.Filter(ResponseType='Bandpass', ...
    PassFreq_bp=[2.395e9 2.405e9], ...   % Absolute: carrier +/- 5 MHz
    RF=2.4e9, SampleRate=100e6);

% WRONG: This creates a filter passing 0-5 MHz absolute (2.4 GHz away from carrier)
filt = rf.Filter(ResponseType='Lowpass', ...
    PassFreq_lp=5e6, RF=2.4e9, SampleRate=100e6);  % Near-zero output!

For a lowpass channel filter of bandwidth B around carrier fc, use ResponseType='Bandpass' with PassFreq_bp=[fc-B/2, fc+B/2] and RF=fc.

Gotcha: RF must be positive -- RF=0 errors with "Value must be positive."

UseFilterOrder

By default (UseFilterOrder=false), filter order is computed automatically from passband/stopband attenuation specs. When UseFilterOrder=true, specify FilterOrder directly and StopFreq_*/StopAtten properties are ignored (warnings emitted if set).

% Auto-order from specs (UseFilterOrder=false, the default)
filt = rf.Filter(ResponseType='Bandpass', ...
    PassFreq_bp=[2.395e9 2.405e9], StopFreq_bp=[2.39e9 2.41e9], ...
    StopAtten=40, RF=2.4e9, SampleRate=100e6);

% Fixed order (UseFilterOrder=true) -- StopFreq/StopAtten ignored
filt = rf.Filter(ResponseType='Bandpass', UseFilterOrder=true, FilterOrder=5, ...
    PassFreq_bp=[2.395e9 2.405e9], RF=2.4e9, SampleRate=100e6);

Properties

Gotcha: Passband/stopband frequency properties change name based on ResponseType: PassFreq_lp, PassFreq_hp, PassFreq_bp, PassFreq_bs. Setting the wrong suffix for the current ResponseType has no effect.

Visualization

visualize(filt);   % Plot filter frequency response

rf.Sparameter -- S-Parameter System Model

Process complex baseband signals through S-parameter data using rational fitting:

Construction

% From Touchstone file (default)
sp = rf.Sparameter(FileName='passive.s2p', CarrierFrequency=2.4e9, SampleRate=10e6);

% From sparameters object
sObj = sparameters('device.s2p');
sp = rf.Sparameter(DataSource='sparameters', SParametersObject=sObj, ...
    CarrierFrequency=2.4e9, SampleRate=10e6);

% From rational fit object
fitObj = rational(sparameters('device.s2p'));
sp = rf.Sparameter(DataSource='rational', RationalObject=fitObj, ...
    CarrierFrequency=2.4e9, SampleRate=10e6);

Example: Process Signal Through Synthetic S-Parameters

freq = linspace(1e9, 4e9, 101)';
s21 = 0.8 * exp(-1j*2*pi*freq*1e-9);  % ~-2 dB, linear phase
sData = zeros(2,2,numel(freq));
sData(1,1,:) = 0.1; sData(2,1,:) = s21;
sData(1,2,:) = s21; sData(2,2,:) = 0.1;
sObj = sparameters(sData, freq);
sp = rf.Sparameter(DataSource='sparameters', SParametersObject=sObj, ...
    CarrierFrequency=2.4e9, SampleRate=10e6);
outSig = sp(randn(1000,1) + 1j*randn(1000,1));
release(sp);

Visualization

visualize(sp);   % Plot S-parameter frequency response

rf.PAmemory -- Power Amplifier with Memory

Model nonlinear power amplifiers with memory effects using Volterra series:

pa = rf.PAmemory;
pa = rf.PAmemory(Model='Memory polynomial', CoefficientMatrix=coeffs);
pa = rf.PAmemory(Model='Cross-term memory', CoefficientMatrix=coeffs);

The coefficient matrix dimensions encode the nonlinear order and memory depth. See the ComputePACoefficientMatrixExample for extracting coefficients from measured data.

rf.MismatchLoss -- Impedance Mismatch

Model frequency-dependent impedance mismatch loss:

ml = rf.MismatchLoss;
ml.MismatchLossData = lossValues;           % dB loss values
ml.MismatchLossFrequency = freqPoints;      % Hz

Inherits from rf.Sparameter. Typically used in idealized baseband simulation chains to model inter-stage mismatch.

Common Patterns

Power Measurement (1-Ohm Convention)

% rf.Amplifier/rf.Mixer use 1-ohm normalized power: P = |V|^2
Vin = sqrt(Pin_W);                     % NOT sqrt(2*50*Pin_W)
Pout_dBm = 10*log10(mean(abs(out).^2)) + 30;

% rfsystem uses 50-ohm: P = |V|^2 / (2*50) -- see reference/system-simulation.md

This differs from rfsystem which uses 50-ohm scaling. Using 50-ohm scaling with rf.Amplifier gives wrong P1dB values (compression 20 dB too early). The 1-ohm convention ensures OIP3/OP1dB specs match the measured compression point.

EVM Measurement

% After amplifier
outNorm = out / mean(abs(out)) * mean(abs(inSig));
evmPct = 100 * rms(outNorm - inSig) / rms(inSig);

Cascading Multiple System Objects

fc = 2.4e9; fs = 100e6;
amp = rf.Amplifier(Gain=15, OIP3=35);
filt = rf.Filter(DesignMethod='Butterworth', ResponseType='Bandpass', ...
    PassFreq_bp=[fc-5e6, fc+5e6], RF=fc, SampleRate=fs);
out1 = amp(inSig);
out2 = filt(out1);

When cascading, ensure SampleRate is consistent across all objects that use it (rf.Filter, rf.Sparameter, and rf.Amplifier/rf.Mixer when noise is enabled). The RF property on rf.Filter must match the carrier frequency of the signal being processed.

For full-system cascaded simulation with automatic Simulink model generation, use rfsystem instead (see reference/system-simulation.md).

Clone for Independent Copies

amp1 = rf.Amplifier(Gain=15, OIP3=35);
amp2 = clone(amp1);   % Independent copy
amp2.Gain = 20;       % Does not affect amp1

Relationship to rfbudget Elements

The rf.* objects process complex baseband signals frame-by-frame. The rfbudget elements define frequency-domain specifications for cascade analysis. Use rfsystem to bridge between them (see reference/system-simulation.md).

Gotchas

1. R2024b minimum -- rf.Amplifier, rf.Mixer, rf.Filter, rf.Sparameter require R2024b+. rf.PAmemory requires R2024a+.

2. No Simulink required -- Unlike rfsystem, the rf.* objects run entirely in MATLAB.

3. SampleRate relevance varies by object -- For rf.Amplifier, only set SampleRate when IncludeNoise=true (emits "not relevant" warning otherwise). For rf.Mixer, only when IncludePhaseNoise=true. But rf.Filter and rf.Sparameter always require SampleRate -- it is a core property for those objects.

4. rf.Filter frequency properties are ABSOLUTE RF frequencies -- PassFreq_bp=[2.395e9 2.405e9] means the passband is 2.395-2.405 GHz absolute. The RF property defines the carrier center for internal baseband conversion. Do NOT use baseband offsets (e.g., PassFreq_lp=5e6 with RF=2.4e9 produces near-zero output). For a channel filter of bandwidth B at carrier fc, use ResponseType='Bandpass' with PassFreq_bp=[fc-B/2, fc+B/2].

5. rf.Filter frequency properties are ResponseType-dependent -- PassFreq_lp for lowpass, PassFreq_bp for bandpass, etc. Wrong suffix emits "not relevant" warning and is silently ignored (defaults remain).

6. rf.Filter RF must be positive -- RF=0 errors with "Value must be positive."

7. rf.Filter UseFilterOrder and stopband specs -- When UseFilterOrder=true (or when it defaults to true), StopFreq_* and StopAtten are "not relevant" and emit warnings. Set UseFilterOrder=false explicitly if you need to specify stopband attenuation goals.

8. Nonlinearity is only for Model='cubic' -- For the default Model='poly', do NOT set Nonlinearity (it emits "not relevant" warning). Poly uses all compression params you provide (e.g., OIP3, OP1dB, OPsat for amplifiers) to fit the polynomial. For Model='cubic', Nonlinearity selects which ONE param shapes the 3rd-order curve. Silent misuse: rf.Amplifier(Gain=15, Model='cubic', OPsat=30) without Nonlinearity='OPsat' applies no compression (Nonlinearity defaults to OIP3 with OIP3=Inf).

9. rf.Mixer Model is not the same as modulator ConverterType -- rf.Mixer uses 'mod'/'demod'/'iqmod'/'iqdemod'. The rfbudget modulator uses ConverterType='Up'/'Down'.

10. rf.Mixer IQ model property relevance is asymmetric -- 'iqmod': RF and PhaseOffset are irrelevant (emit warnings). LO does not warn and does not affect frequency conversion, but scales phase noise when IncludePhaseNoise=true. 'iqdemod': RF is relevant, but LO and PhaseOffset are not. Setting irrelevant properties emits "not relevant" warnings.

11. Properties locked during use -- Nontunable properties cannot be changed between step calls. Call release(obj) before modifying.

12. Input must be column vectors or matrices -- Row vectors error. Each column is an independent channel.

13. rf.Sparameter default file is unitygain.s2p -- This is a built-in identity file. Always set FileName or SParametersObject explicitly.

14. rf.Amplifier/rf.Mixer use 1-ohm normalized power -- Vin = sqrt(Pin_W) and Pout = |Vout|^2. This differs from rfsystem which uses 50-ohm (Vin = sqrt(2*50*Pin_W)). Using 50-ohm scaling with rf.Amplifier gives wrong P1dB values (compression 20 dB too early). The 1-ohm convention ensures OIP3/OP1dB specs match the measured compression point.

Conventions

• Use tiledlayout/nexttile for multi-panel signal analysis plots

• Always label axes with units (dBm, MHz, degrees) and include figure titles

• Call release(obj) before modifying Nontunable properties

• Use visualize() or visualizePower()/visualizePhaseNoise() to inspect component behavior before simulation

Copyright 2026 The MathWorks, Inc.