I/Q & the Complex Envelope

01 / 10

The sideband that moved to the wrong side

How can the same stored I/Q tone appear above the carrier in one reconstruction and below it in another?

The recurring 20.0 kbit/s condition-monitoring link now needs an executable handoff from a normalized waveform description to a real RF signal. A transmitter model reports a +50.0 kHz complex tone. One bench produces 2.450050 GHz; another produces 2.449950 GHz. Neither plot is self-explanatory: their Q-mixer signs differ.

One coordinate record, three linked views, one convention-dependent RF result

D2 · I/Q plane

D2 · signed spectrum

R0 · positive RF half

Arrows and line heights are explanatory normalized graphics, not measured data.

Think about itIf u(t) rotates counterclockwise at +50.0 kHz, must its RF line be above the carrier?

Answer

Only after the RF reconstruction sign is stated. With sRF = I cos − Q sin, yes. With sRF = I cos + Q sin, the same stored samples map below the carrier.

Common misconceptionPositive Q always means the upper sideband.

Q is a coordinate. Sideband placement follows from the complete exponential, transform, and mixer-sign convention—not from the letter Q alone.

Evidence Complex-exponential modulation and Fourier translation: P02-S1 and P02-S3; sideband placement is derived explicitly in sections 4–5.

02 / 10

Why a real RF signal has redundant spectral halves

What does a complex envelope keep that a single real baseband waveform cannot?

For real sRF(t), its transform obeys S(−f) = S*(f). The negative-frequency half is required, but once the positive half is known it supplies no independent information. A real cosine therefore contains equal phasor coefficients at +f and −f.

$A \cos (2\pi f_{0}t + \phi) = (\frac{A}{2})e^{j\phi}e^{j 2\pi f_{0}t} + (\frac{A}{2})e^{-j\phi}e^{-j 2\pi f_{0}t}$ The two coefficients are conjugates; neither half is an independently selectable sideband for a real waveform.

The analytic signal s_a(t) = sRF(t) + jℋ{sRF(t)} removes the redundant negative-frequency half and doubles non-DC positive-frequency coefficients. Removing a known carrier gives the complex envelope u(t) = s_a(t)e^−j2πfct. Reconstruction returns the physical real waveform: sRF(t) = Re{u(t)e^j2πfct}.

Related representations are not interchangeable arrays
Representation	Independent spectrum	Reference	Typical scale
Real RF sRF(t)	Positive and conjugate negative halves	Absolute RF frequency	R0 volts only with load and scale
Analytic signal sa(t)	Nonnegative frequencies	Absolute RF frequency	Complex voltage-like quantity if calibrated
Complex envelope u(t)	Signed baseband frequencies	Rotating frame at fc	D2/D3 dimensionless by default

Go deeperWhy does the analytic signal double the positive half?

Taking Re{·} divides a complex exponential into two half-amplitude conjugate terms. The analytic signal retains one term at its full amplitude so that taking the real part later reconstructs the original waveform, not one half of it.

Evidence Analytic-signal and transform definitions: P02-S1, Chapters 3–5; P02-S2, Chapter 2.

03 / 10

I and Q are orthogonal coordinates

How can two real numbers carry one amplitude and one phase without ambiguity?

Write u(t) = I(t) + jQ(t) = A(t)e^jφ(t). I and Q are Cartesian coordinates on orthogonal axes; A and φ are the same point in polar form. Use A = √(I² + Q²) and φ = atan2(Q, I). A one-argument arctangent loses quadrant information and fails at I = 0.

Cartesian: I = A cos φ
Q = A sin φ
Polar: A = √(I² + Q²)
φ = atan2(Q, I)
Power-like quantity: |u|² = I² + Q²
dimensionless until scaled
At A = 0: φ is undefined
do not force it to 0°

Common misconceptionI is amplitude and Q is phase.

Both coordinates jointly determine both amplitude and phase. Neither axis owns one polar quantity.

Scale boundary. D2 and D3 values are dimensionless unless a model declares otherwise. Calling A “volts” or |u|² “watts” requires a gain definition, peak/RMS convention, impedance, and reference plane.

Evidence Cartesian/polar complex representation: P02-S1 and P02-S2; plane and normalization language follows the approved Path 02 portfolio contract.

04 / 10

Reconstruct RF under one explicit sign convention

What real waveform does one I/Q point produce?

$s_{\mathrm{RF}}(t) = \operatorname{Re}{u(t)e^{j 2\pi f_c t}} = I(t)\cos (2\pi f_c t) - Q(t)\sin (2\pi f_c t)$ Primary convention for this lesson: positive complex baseband frequency maps above the carrier.

Substituting I = A cos φ and Q = A sin φ gives sRF(t) = A cos(2πfct + φ). This is why atan2(Q, I) is the carrier phase under the declared minus-Q convention.

Checked normalized points under sRF = I cos − Q sin
(I, Q)	A	φ = atan2(Q, I)	Real RF waveform
(+1, 0)	1	0°	+cos(2πfct)
(0, +1)	1	+90°	−sin(2πfct)
(−1, 0)	1	+180°	−cos(2πfct)
(0, −1)	1	−90°	+sin(2πfct)
(1/√2, 1/√2)	1	+45°	cos(2πfct + 45°)

Worked point: (I, Q) = (1/√2, 1/√2)

A = 1, φ = +45°, and sRF(t) = cos(2πfct + 45°). The statement is normalized; it is not yet a one-volt-peak claim.

Think about itIf Q alone changes sign while I is unchanged, what happens to this point and its RF phase?

Answer

The point reflects across the I axis, u becomes u*, and +45° becomes −45°. For a rotating sequence, this conjugation reverses spectral orientation.

Evidence The table follows by direct substitution into the displayed reconstruction equation and is regression tested in iq-reconstruction/1.0.

05 / 10

Upconversion and spectral translation

How do the I and Q mixer products cancel one image and reinforce the other?

Let u(t) = e^j2πfbt, so I = cos(2πfbt) and Q = sin(2πfbt). Under the primary convention, product-to-sum identities make the lower terms cancel:

$\cos (2\pi f_b t)\cos (2\pi f_c t) - \sin (2\pi f_b t)\sin (2\pi f_c t) = \cos [2\pi (f_{c} + f_{b})t]$ For fb > 0, the desired real RF line is at fc + fb; conjugate symmetry also supplies the required line at −(fc + fb).

If the hardware instead computes I cos + Q sin, the identity becomes cos[2π(fc − fb)t]. That convention is valid—but only if every generator, file, simulation, and receiver agrees.

Minus-Q convention+fb → fc + fbupper translation for positive complex frequency

Plus-Q convention+fb → fc − fblower translation for the same stored tone

Common misconceptionA real cosine mixer can make a one-sided translation by itself.

A single real multiplication creates translated copies on both sides. Orthogonal I/Q products create controllable cancellation of one image.

Evidence Fourier modulation property and quadrature mixing: P02-S1, P02-S2, and P02-S3; expansion checked numerically against time-domain correlation.

06 / 10

Downconversion, filtering, and scaling

Why do ideal receiver equations contain a factor of two and an explicit low-pass filter?

Mix the real RF waveform with coherent cosine and negative-sine references:

2sRF cos θc= I + I cos 2θc − Q sin 2θcLPF → I

−2sRF sin θc= Q − I sin 2θc − Q cos 2θcLPF → Q

The factor two compensates the one-half mixer product. Without it, an ideal low-pass result is I/2 and Q/2. A practical receiver can absorb that factor into calibrated gain, but the scale must be stated. Phase or frequency error in the local oscillator rotates or continuously spins the recovered complex envelope.

D2ideal u[n]dimensionless
D3scaled digital I/Qdimensionless unless declared
A0DAC/ADC boundarysample and reconstruction assumptions
A1analog I/Q portsvolts only with scale
R0RF portreal waveform, load required
R3receiver detectorgain and bandwidth declared

Ideal boundary. The equations assume matched LO frequency and phase, ideal multipliers, and an LPF that passes the full envelope while rejecting the 2fc terms. Filter transition, group delay, noise, ADC behavior, and synchronization are later design work.

Evidence Coherent quadrature detection follows P02-S2, Chapters 2 and 4; every mixer factor is retained in the displayed algebra.

07 / 10

Magnitude, phase, and instantaneous frequency

Which I/Q-derived quantities remain meaningful along a changing trajectory?

Envelope magnitudeA(t) = |u(t)|

Distance from the origin. Convert to calibrated voltage or power only after scale and impedance are defined.

Wrapped phaseφ(t) = atan2(Q, I)

Correct quadrant, normally reported in (−π, π]. Undefined when A = 0.

Instantaneous offsetfi(t) = (1/2π)dφu/dt

Requires a consistent unwrapped, differentiable phase and nonzero magnitude.

Complex power proxy|u(t)|² = I² + Q²

Preserved by conjugation, so it cannot by itself detect spectral inversion.

At a deep envelope null, phase noise and numerical roundoff can produce a violent angle jump; differentiating it yields a large number with little physical meaning. A robust implementation carries a magnitude-validity mask and documents unwrap and derivative methods.

Go deeperPeak, RMS, and complex-envelope scaling

Under the displayed reconstruction, a constant |u| = A produces a real sinusoid with peak amplitude A in the same voltage-like scale. Its RMS value is A/√2 and its average power into a real resistance R is A²/(2R). Those are valid only when A is explicitly a peak voltage at the declared R0 plane.

Evidence Magnitude, phase, and complex-envelope definitions: P02-S1 and P02-S2; scale discipline follows the Path 02 reference-plane contract.

08 / 10

Spectral inversion and metadata failures

How can a record keep exactly the same sample power yet represent the opposite spectrum?

Conjugating u = I + jQ gives u* = I − jQ. Every sample keeps |u|², but a tone at +fb becomes a tone at −fb. A power check passes while upper and lower spectral content exchange. This makes convention metadata part of the signal, not optional documentation.

$\begin{aligned}u(t)&=e^{j2\pi f_bt}\Rightarrow u^*(t)=e^{-j2\pi f_bt}\\|u(t)|^2&=|u^*(t)|^2=1\end{aligned}$ Q polarity reversal is a spectral-orientation change, not a harmless cable-label edit.

Minimum metadata before exchanging an I/Q dataset
Field	This lesson's declaration	Failure signature
Complex definition	u = I + jQ	Swapping I/Q or changing Q polarity conjugates or rotates the record.
RF reconstruction	Re{u exp(j2πfct)}	Equivalent to I cos − Q sin; a plus-Q mixer maps positive baseband below fc.
Transform sign	X(f) = ∫x(t)exp(−j2πft)dt	Changing the exponent reverses the frequency-axis interpretation.
Frequency order	Signed, increasing −fs/2 … +fs/2	Unshifted FFT arrays can make the wrapped negative half look positive.
Units and scale	D2/D3 dimensionless unless declared	Volts, RMS power, and watts require gain, impedance, and reference plane.
Time/sample reference	t = 0 and fs stated	An unknown delay rotates phase; a wrong fs rescales frequency.

Think about itA file's total I² + Q² matches the source, but a positive calibration tone lands at −50 kHz. Is the file validated?

Answer

No. The energy invariant passed, but the orientation invariant failed. Audit Q polarity, channel order, transform sign, frequency order, and any mixing stage before acceptance.

Evidence Conjugation and Fourier-sign behavior: P02-S1; dataset contract fields are derived from the approved Path 02 portfolio requirements.

09 / 10

I/Q imbalance, DC offset, and LO leakage

Which visible errors point to branch mismatch, and which point to DC?

Let g be the Q-to-I voltage-gain ratio and let ϕ be quadrature deviation from 90°. For a complex calibration tone, the ideal desired coefficient is replaced by (1 + ge^jϕ)/2 and the image coefficient by (1 − ge^jϕ)/2, apart from common gain.

$\mathrm{IRR} = \frac{1 + g^{2} + 2g \cos \varphi}{1 + g^{2} - 2g \cos \varphi}$ Power ratio. g is a linear voltage-gain ratio; ϕ is the deviation from perfect quadrature, not the total 90° separation.

Checked imbalance case

For +1.0 dB relative Q gain, g = 10^1/20 = 1.122018. With ϕ = 3.0°, IRR = 250.496, or 23.988 dB. Ideal g = 1 and ϕ = 0 gives ∞ in the mathematical model—not a numerical floor.

Cause, visible symptom, and isolating observation
Impairment	First-order mechanism	Signature	Useful check
Common I/Q gain	Scales desired, image, and DC terms together	Absolute amplitude change; tone IRR unchanged	Known A1/R0 scale or calibrated loopback
Q gain relative to I	Incomplete sideband cancellation	Mirror image; constellation stretched	Inject a complex tone and compare desired/image lines
Quadrature phase error	LO branches differ from 90°	Mirror image; constellation skew/rotation	Fit desired/image coefficients with sign convention fixed
I/Q DC offset	DC mixes with the quadrature LOs	Line at fc (LO leakage); constellation displaced	Remove mean at D3 or calibrate offsets at A1

Decision lab · iq-reconstruction/1.0

I/Q Reconstruction Bench

Trace one deterministic D2 complex-envelope input through D3 scaling, A0 quadrature reconstruction, and the R0 real-RF result. Tone metrics use exact analytic line coefficients.

Input

RF reconstruction convention

Point phase, φ = atan2(Q, I)

Visible evidence

Demo carrier1.000 MHz

Signed baseband offset50.0 kHz

I branch gain0.0 dB

Q gain relative to I0.0 dB

Quadrature error, φ0.0°

I DC offset at D30.00

Q DC offset at D30.00

D2 · normalized complex envelope

I/Q plane and reconstruction path

D2u[n]dimensionless
D3gain + DCdimensionless
A0DAC outputscale required
A1I/Q portsvolts if declared
R0real RFload required

R0 · normalized voltage-like waveform

Four carrier cycles in time

The trace is deterministic and sampled from the displayed equation. No watt or volt claim is made until an A1/R0 scale, termination, and peak-or-RMS rule are declared.

D2 signed frequency → R0 absolute RF

Tone translation and imperfect cancellation

Signed complex baseband

Positive absolute RF frequencies · conjugate negative-frequency half not drawn

Desired
Image
LO leakage

Exact tone coefficients · normalized amplitudes

Desired, image, and carrier evidence

Desired RF line: 1.050 MHz
Image RF line: 950.0 kHz
IRR, power ratio: ∞ (ideal model)
Carrier leakage: −∞ (ideal model)

Image: No image in the ideal mathematical model.

LO leakage: No carrier leakage in the ideal mathematical model.

Common I gain: changes absolute normalized amplitude, not ideal tone IRR.

Component	Frequency	Normalized amplitude	Relative to desired
Desired	1.050 MHz	1.000000	0.00 dB
Image	950.0 kHz	0.000000	−∞ (ideal model)
LO leakage	1.000 MHz	0.000000	−∞ (ideal model)

s(t) = 1.0000[I(t) + 0.00]cos(2πfᶜt) − 1.0000[Q(t) + 0.00]sin(2πfᶜt + 0.0°)Current D2 input: I = 1.0000, Q = 0.0000, A = 1.0000, φ = 0.0°. Branch-weighted D3 point: (1.0000, 0.0000).

Scope: ideal mathematical mixers and an exact tone-line impairment model. PRBS/QPSK is a deterministic trajectory preview; its occupied spectrum requires pulse shape, record, and estimator definitions covered later. Numerical plotting never turns ideal ∞ rejection into a physical floor.

The default is a +50.0 kHz unit complex tone at D2, reconstructed around 1.000 MHz with s(t) = I cos(2πfᶜt) − Q sin(2πfᶜt). It produces the desired line at 1.050 MHz, zero ideal image, and no carrier leakage. The static worked examples below preserve the same result without JavaScript.

Evidence Coefficients and IRR are independently derived and regression tested in iq-reconstruction/1.0; qualitative constellation signatures are cross-checked against P02-S5.

10 / 10

Validate the I/Q contract

What must waveform decision record v3 contain before another engineer can reconstruct the intended RF?

Version 3 does not choose the final symbol mapping or sample rate. It closes the coordinate and reconstruction contract so those later decisions cannot silently reverse a sideband or invent a voltage scale.

Complex definition · D2: u[n] = I[n] + jQ[n], with I first and positive Q explicitly declared.
RF reconstruction · A1 → R0: sRF(t) = Re{u(t)e^j2πfct} = I cos − Q sin.
Frequency convention: X(f) uses the negative exponential; signed D2 order is −fs/2 … +fs/2 after shift.
Sideband check: A +50.0 kHz unit tone must reconstruct at fc + 50.0 kHz under the chosen convention.
Scale and units: D2/D3 are dimensionless normalized samples. A1/R0 volts require later peak/RMS gain and impedance declarations.
Reference planes: D2 ideal source, D3 digital scaling, A0 conversion boundary, A1 analog I/Q, R0 transmit RF, R3 receive detector.
Known invariants: Canonical-point RF phases, Cartesian/polar round-trip, orientation tone, and mean |u|² under conjugation.
Open work: Sampling, pulse shape, analog filtering, synchronization, noise, calibration limits, and compliance remain unresolved.

Dataset acceptance test

Verify channel order, numeric format, sample rate, center frequency, time origin, scale, and reference plane.
Inject or locate a signed calibration tone and confirm its baseband orientation and reconstructed RF side.
Check canonical points and |u|², then preserve the exact metadata with the samples.

Reject: “interleaved I/Q, 1 MS/s” without channel order, Q polarity, sign convention, numeric scale, center frequency, or plane. The bytes may be readable; the waveform is not reproducible.

Symptom → first audit → likely owner
Observed symptom	First audit	Likely next owner
Spectrum mirrored around fc	Q polarity, I/Q order, mixer sign, FFT order	Waveform / converter interface
Symmetric image grows	Relative gain and quadrature error	Modulator calibration
Line exactly at fc	I/Q DC and LO feedthrough	Converter / RF hardware
Amplitude off by √2 or 2	Peak/RMS and mixer scaling	Model / measurement contract
Phase spins with time	Carrier-frequency reference	Synchronization design

Owned later. Real/complex sampling and alias plans belong to 02.4; constellation mapping to 02.5; pulse shaping to 02.6; EVM and calibrated impairment comparison to 02.7. This module supplies their common I/Q contract.

Handoff to 02.4

The RF reconstruction is now unambiguous. The next decision is how a continuous complex envelope becomes a finite-rate digital record without aliasing, clipping, or hidden converter assumptions.

Ungraded review

Check your understanding

Answer each question in your own words, then reveal the model answer.

01Under sRF(t) = I(t)cos(2πfct) − Q(t)sin(2πfct), where does u(t) = exp(j2πfbt) appear for fb > 0?
Model answer
At fc + fb. Expanding I = cos(2πfbt) and Q = sin(2πfbt) makes the fc − fb terms cancel and the fc + fb terms add. The answer reverses for the plus reconstruction convention.
02Why is the negative-frequency half of a real RF spectrum called redundant rather than nonexistent?
Model answer
A real waveform requires conjugate symmetry, S(−f) = S*(f). The negative half is physically required by the real representation, but it carries no independent degrees of freedom once the positive half is known.
03What RF waveform does the normalized point I = 0, Q = +1 produce under the minus convention?
Model answer
sRF(t) = −sin(2πfct) = cos(2πfct + 90°). Its normalized magnitude is one and atan2(Q, I) gives +90°.
04What must be true before dφ(t)/dt is interpreted as instantaneous frequency?
Model answer
The phase must be consistently unwrapped and differentiable over the interval, the complex envelope magnitude must be nonzero, and the time reference and sample rate must be known.
05A positive-frequency complex tone appears below the carrier. What should you audit before blaming the RF hardware?
Model answer
Audit the reconstruction sign, FFT sign and frequency order, I/Q channel order, Q polarity, mixer-side convention, and metadata. Any one can conjugate or invert the representation while leaving mean complex power unchanged.
06Which first-order impairments create an image and which create a carrier line in this model?
Model answer
Relative I/Q gain error and quadrature phase error prevent image cancellation. I/Q DC offsets mix to fc and create LO leakage. A common gain changes absolute scale but not the ideal tone image-rejection ratio.

Sources and further study

The lesson cites and paraphrases these sources. The linked MIT OpenCourseWare sequence and Rohde & Schwarz application note were available when checked on 5 September 2026. All plots here are derived or simulated teaching evidence; none is a measured trace.

Primary technical references

P02-S1 · Informative: A. V. Oppenheim, A. S. Willsky, and S. H. Nawab, Signals and Systems, 2nd ed., Chapters 3–7.
P02-S2 · Informative: J. G. Proakis and M. Salehi, Digital Communications, 5th ed., Chapters 2, 4, and 5.
P02-S3 · Informative: B. Sklar, Digital Communications: Fundamentals and Applications, 2nd ed., Chapters 1–4.
P02-S4 · Informative: MIT OpenCourseWare: Unified Engineering Signals and Systems, especially S15–S18.
P02-S5 · Orientation only: Rohde & Schwarz, Understanding Error Vector Magnitude, Version 01.00, October 2022.

Evidence and scope

Defined: transform sign, signed frequency, u = I + jQ, RF reconstruction sign, units, and reference planes are stated before use.
Derived: canonical RF phases, tone translation, coherent downconversion, mismatch coefficients, and IRR follow the shown equations.
Simulated: iq-reconstruction/1.0 is deterministic and checks analytic coefficients against direct time-domain correlation.
Illustrative: the 20.0 kbit/s condition-monitoring link and 2.450 GHz carrier name no wireless standard or compliance claim.

Previous in Signals & ModulationWhy Modulation Exists

Next in Signals & Modulation · plannedSampling & Digital Representation

Back to Signals & Modulation