35-year-old 3-way speaker: crossover for tube amplification
35-year-old 3-way speaker
Electroacoustic evaluation, system geometry, directivity and listening positioning
This report analyses a three-way speaker about 35 years old and the five-component crossover proposed for it. The entire system —the three transducers, the filter and the baffle geometry— has been simulated in LspCAD 6.20 pro before building the physical cabinet. The starting data are the previous measurements of the speakers: a 27 cm woofer with paper/mica cone, a 7.7 cm midwoofer and a 5 cm tweeter, both paper. The simulation has been frozen as Reference Model MR-A-01. From here on, the report distinguishes three things that should not be confused: what the model predicts, what is calculated with formulas and what has not yet been measured. That distinction is the backbone of the document.
What it is. A three-way speaker has three drivers inside (one for bass, one for midrange and one for treble) and an electrical filter —the crossover— that distributes the audio signal among them. This report studies a 35-year-old speaker and proposes a new five-component crossover, all simulated in LspCAD 6.20 pro before building the cabinet.
The central unknown. The simulation says the system should respond fairly flat (±1 dB between 400 Hz and 4 kHz), present a minimum impedance of about 4.9 Ω and show two vertical notches consistent with physics. But all of that are model predictions, not measurements. The question that organises the project is: will these figures hold when the speaker exists as a real object and can be measured?
What is not known. Real sensitivity, real impedance, distortion (THD), compression, baffle step, real in-room response... None of that has been measured. Even the "acoustic offsets" of the three drivers are values the model has adopted, not verified data on the physical unit.
How it is decided. The plan is orderly: freeze the model (already done: MR-A-01) → build the cabinet exactly as simulated → measure impedance, response and THD → compare with the recorded predictions → only then decide whether the crossover needs changes. The next step is not to optimise the simulation further: it is to build and measure MR-A-01.
Technical glossary · sixteen terms in one line
PART I — 35-YEAR-OLD 3-WAY SPEAKER · CROSSOVER A
This report analyses a three-way speaker and the crossover proposed for it. It is not a report on a commercial product: it is the study of one specific system, Crossover A, on a 35-year-old speaker, simulated in LspCAD 6.20 pro and frozen as Reference Model MR-A-01 before building the physical cabinet. The distinction between prediction, calculation and measurement is the backbone of the document.
GETTING STARTED — WITH NO PRIOR KNOWLEDGE (G1–G5)
GETTING STARTED — WITH NO PRIOR KNOWLEDGE
Before entering the report, it is worth having five ideas clear. We present them here with numerical examples. They will be used throughout the text without further explanation.
Basic guide
Before starting. This report analyses a three-way speaker and the filter that distributes the signal among its drivers. You do not need to be a sound engineer to follow it, but it helps to have five basic ideas clear. Let's look at them one by one, with numerical examples. Throughout the report they will be used without further explanation.
G1. What a three-way speaker is and why it needs a crossover
A three-way speaker shares the work among three transducers. The woofer is large and moves a lot of air: it handles the bass. The tweeter is small and light: it handles the treble. The midwoofer sits in the middle and covers the central part of the spectrum, which is where the human ear is most sensitive.
None of the three reproduces the whole spectrum well on its own. The crossover is the electrical filter that does that sharing. It sits between the audio signal and the three transducers, and each one receives only the part of the spectrum that corresponds to it:
Audio signal → Crossover → Woofer (bass)
→ Midwoofer (midrange)
→ Tweeter (treble)Crossover A uses a parallel topology: the three ways hang from the same input point, each with its own filter. The five components are distributed as follows:
- Woofer: L3 (1.8 mH) in series with the driver → first-order electrical low-pass filter (6 dB/octave asymptotically in the ideal case).
- Midwoofer: C1 (33 µF) + R2 (2.2 Ω) in series with the driver, L4 (0.82 mH) in shunt → second-order high-pass electrical network (12 dB/octave).
- Tweeter: C2 (2.7 µF) in series with the driver → first-order electrical high-pass filter (6 dB/octave asymptotically in the ideal case), with no attenuation resistor.
The order of a filter is the number of reactive components (coils and capacitors) in the signal path. In ideal networks, a higher order allows a steeper asymptotic slope, but it does not determine the acoustic slope or a fixed phase shift at crossover: driver impedance, natural response and interaction between ways also matter. As we will see, Crossover A mixes the two: first order in the woofer and tweeter, second order in the midwoofer.
G2. Which data is reliable and which is not
Each statement in the report carries a label indicating its level of evidence. The rule is simple: no statement rises above the category of its source evidence. If a data point comes from a simulation, it is a prediction; if it comes from a formula, it is a calculation; if it comes from a physical measurement, it is a measurement.
This matters because the report distinguishes three levels for any numerical value of the system: prediction (what the LspCAD model says), validation (what the physical measurement says) and specification (what ends up characterised for the finished cabinet). You do not go from level 1 to level 3 without measuring. The distance between the two is the measurement, and it cannot be skipped without losing traceability.
G3. Frequency response, ripple and directivity
The frequency response indicates how the sound level (SPL, measured in decibels) varies as the frequency changes. If the level is approximately constant, we say the response is "flat". Ripple is the maximum deviation from the average in a specific band: a ripple of ±1 dB means that the curve oscillates 2 dB peak-to-peak around its average value.
Directivity, on the other hand, describes how sound is distributed in space. In a three-way system this is especially critical, because each transducer has its own radiation pattern and interference between drivers produces notches (deep, narrow level drops) at specific angles.
To understand notches, it helps to remember what wavelength is. It is the distance a wave travels in one cycle, and it depends on frequency according to the formula:
Where c is the speed of sound (about 343 m/s in air at room temperature) and f is the frequency. At 4.5 kHz —the crossover frequency between midwoofer and tweeter— the wavelength is about 76 mm. For two coherent sources in phase on the reference axis, a half-wavelength path difference (about 38 mm) can cause cancellation if their amplitudes are comparable. The real result depends on the phase and level of each way. In other words: small distances, at high frequencies, produce significant interference.
G4. Impedance, EPDR and the relationship with the amplifier
A "6 Ω" speaker does not present 6 Ω at all frequencies. Its impedance varies: it can drop to 4.9 Ω in one zone and rise to 16 Ω in another. That variation interacts with the amplifier's output impedance (Zout) as a voltage divider:
If Zout is very low, the voltage reaching the speaker is almost the same at all frequencies and the response remains flat. If Zout is high, the voltage received by the speaker follows the impedance curve, and that translates into coloration. EPDR (Equivalent Peak Dissipation Resistance) is an indicator of how much current the speaker-crossover combination actually demands in the worst case. The lower it is, the more demanding the load is for the amplifier.
G5. Check exercises
To fix the ideas, here are four simple calculations. The solutions are at the end, in case you want to try before looking:
- A 1.8 mH coil over a 6 Ω load: at what frequency is its electrical crossover?
- What is the wavelength at 4.5 kHz?
- With 89 dB/W/m and 3 m distance, what SPL does 1 W produce?
- If the ripple in a band is ±1 dB, what does that mean peak-to-peak?
See solutions
- f = Z / (2πL) = 6 / (2π · 0.0018) ≈ 531 Hz
- λ = 343 / 4500 ≈ 76 mm
- SPL = 89 − 20·log₁₀(3) ≈ 79.46 dB
- That the curve oscillates 2 dB between its minimum and maximum (±1 dB with respect to the average) in that band.
0. Executive summary
Twenty-second read
Topology. Five components distributed across three parallel ways. The woofer has a series coil (L3, 1.8 mH) that lets only low frequencies through. The midwoofer has a capacitor (C1, 33 µF) and a resistor (R2, 2.2 Ω) in series, plus a parallel coil (L4, 0.82 mH); the set behaves as a second-order filter that lets the midrange through. The tweeter has only a series capacitor (C2, 2.7 µF) and no deliberate attenuation resistor. CONFIRMED
Geometry. Flat, vertical baffle (90° to the floor), with the three drivers mounted flush. In the model, three acoustic offsets were adopted: +0 mm for the tweeter, −5 mm for the midwoofer and −20 mm for the woofer. They are included in the simulation. PHYSICAL MOUNTING: CONFIRMED · ACOUSTIC OFFSETS: ADOPTED IN MODEL, PENDING MEASUREMENT
Ripple (model). Between 400 Hz and 4 kHz, the model predicts a ripple of ±0.7 to ±1 dB. VISUAL ESTIMATE Between 4 kHz and 10 kHz the ripple rises to ±1.5–2 dB. VISUAL ESTIMATE PENDING MEASUREMENT
Directivity (model). Horizontally, ~2 dB at ±20° and ~4–6 dB at ±30° in 6–10 kHz. Vertically, two notches consistent with physics: one at ~1 kHz at ±60–70° (woofer-mid) and another at 4.5 kHz at ±25–30° (mid-tweeter). LspCAD PREDICTION
Impedance (model). Provisional minimum close to 4.9 Ω at ~1.4 kHz, with moderate phase. No simultaneous combination of very low impedance and high phase. LspCAD PREDICTION PENDING MEASUREMENT
Listening. Ear at the tweeter-midwoofer junction, speakers at 20–25° per side, 2.5–3.5 m distance, slight toe-in. INFERENCE FROM THE MODEL
Project status
In a project like this it is worth separating three different questions, which are sometimes mixed. Object status: what it is, whether it is closed or can be modified. Prediction status: what the model said, with what degree of commitment. Validation status: what the measurement said, or whether there is not one yet.
| Element | Object status | Prediction status | Validation status |
|---|---|---|---|
| Crossover components (L3, C1, R2, L4, C2) | FROZEN | — | — |
| Woofer–mid crossover | PREDICTION ≈700–900 Hz | PENDING | |
| Mid–tweeter crossover | PREDICTION ≈4–5 kHz | PENDING | |
| Axial response | PREDICTION | PENDING | |
| Directivity (horizontal and vertical) | PREDICTION | PENDING | |
| Impedance and phase | PREDICTION | PENDING | |
| EPDR | MODEL ESTIMATE | PENDING | |
| Sensitivity | CALCULATION HYPOTHESIS: 89 dB/W/m | PENDING | |
| Baffle step | INCOMPLETE MODEL | PENDING | |
| THD and driver compression | ND — outside the model's scope | NOT APPLICABLE TO MODEL |
FROZEN means that the value should not be modified before obtaining physical measurements. It does not mean it is optimal. ND means that the model does not contain the phenomenon and therefore does not issue a prediction about it.
1. Crossover A configuration
In this section we will see how the crossover is built. We start from a 3-way speaker with paper/mica and paper cones, about 35 years old. The consolidated design consists of five components distributed across three parallel ways. Each way feeds a different transducer and filters the signal independently.
An important particularity: the tweeter operates without a deliberate series attenuation resistor. That means its relative level is not adjusted with a resistor, but is determined by the sensitivity of the transducer itself, by C2 and by the overall response of the system.
| Component | Value | Way | Function | Evidence |
|---|---|---|---|---|
| L3 | 1.80 mH | Woofer | 1st-order low-pass (in series with the driver) | CONFIRMED |
| C1 | 33.0 µF | Midwoofer | High-pass (in series with the driver) | CONFIRMED |
| R2 | 2.2 Ω | Midwoofer | Damping/level (in series with C1) | CONFIRMED |
| L4 | 0.82 mH | Midwoofer | Shunt, in parallel with the driver | CONFIRMED |
| C2 | 2.70 µF | Tweeter | 1st-order high-pass (in series), without R1 | CONFIRMED |
Midwoofer topology
The combination C1 + R2 in series and L4 in shunt forms, as a whole, the second-order high-pass electrical network of the midrange way. It should not be broken down as "two independent high-passes that add up", nor should L4 be described as an autonomous ground shunt of the driver. The three components act as a team, not as loose pieces.
- C1 and R2 are in series with the driver. At low frequency, C1 presents a high impedance and limits the current reaching the midwoofer.
- L4 is in shunt (in parallel) with respect to the driver. At low frequency, its impedance is low and it reduces the voltage applied to the transducer.
- As frequency increases, C1 reduces its impedance and L4 increases its own, and that progressively allows excitation of the midwoofer.
Unlike the woofer (which does have a real electrical low-pass thanks to L3) and the tweeter (which does have a real electrical high-pass thanks to C2), the midwoofer has no electrical low-pass. In the model used, its upper cutoff is determined mainly by the natural rolloff of the driver itself. The starting .frd shows this clearly: about 87.6 dB at 5 kHz, 83.3 dB at 6 kHz and 78.0 dB at 10 kHz. READING OF THE STARTING FILE DESIGN INFERENCE
Decision taken: Crossover A is frozen as an experimental reference. The next leap in knowledge must come from the physical cabinet, not from continuing to move values in the simulation. INFERENCE
Why nominal Z and acoustic crossover do not coincide
A quick check with the nominal impedance of 6 Ω can give the impression that the filter values do not produce the expected crossover. Let's see what comes out:
However, LspCAD does not calculate with a fixed nominal Z, but with the starting .zma files, which reflect the real impedance of each transducer frequency by frequency. And that changes things:
- The midwoofer cutoff towards 700–900 Hz is defined by the combination of C1 and L4, not by an isolated component.
- The tweeter cutoff towards 4–5 kHz does not depend only on C2. With the model impedance (~7–8 Ω around 5 kHz), C2 = 2.70 µF gives an electrical reference frequency of ~7.4–8.4 kHz. The acoustic transition starts earlier because the tweeter's own response already rises from ~2 kHz.
That is why the acoustic crossover of the model (700–900 Hz between woofer and midwoofer, 4–5 kHz between midwoofer and tweeter) is coherent with the simulation, even though the calculations made with nominal Z give different figures. It is a good reminder: in a real filter, simple formulas only orient. LspCAD PREDICTION
Component selection — specific specifications
If you are going to build Crossover A, you will need these five components. Here are the verifiable electrical specifications of each. Supplier part codes are not cited because they change between catalogues and countries; any distributor can locate the product by these technical references.
| Component | Specification | DCR / Note |
|---|---|---|
| L3 (1.80 mH) | Preferably air core, 1.4 mm Ø wire (AWG15) — Jantzen Air Core or equivalent | DCR ≈0.42–0.43 Ω. Alternative: 14 AWG foil, DCR ≈0.36 Ω. The coil DCR is part of the series impedance of the bass way and modifies the level and Q of the filter; its real resistance should be introduced into the simulation. Core choice should be assessed for DCR, saturation, linearity and cost. A ferromagnetic core may even offer lower DCR; do not substitute without updating the model. |
| L4 (0.82 mH) | Air core, 1.6 mm Ø wire (AWG14) — Jantzen Air Core or equivalent | DCR ≈0.21 Ω. Alternative: 1.2 mm Ø, DCR ≈0.34 Ω. In shunt; its DCR does not degrade the woofer damping, but it sets the high-pass slope of the midwoofer. |
| R2 (2.2 Ω) | Non-inductive, 12–15 W (Mundorf MResist Supreme 15W, Mills MRA-12, Jantzen Superes 12W) | Actual dissipation in this branch has not been calculated. The suggested 12–15 W rating is provisional and must be checked against simulated RMS current and dissipation, thermal margins and ventilation. |
| C1 (33.0 µF) | MKP 250 V or higher. Mundorf MCAP EVO, Jantzen Cross Cap 400 V, ClarityCap CSA, Audyn Cap Plus. | If exact 33 µF cannot be found: 2× 15 µF + 2× 1.5 µF in parallel, same brand. Avoid bipolar electrolytic: ESR and non-linearity affect the low-midrange. |
| C2 (2.70 µF) | MKP 250 V or higher. Mundorf MCAP EVO Oil 450 V, Jantzen Superior Z-Cap 800 V, ClarityCap CSA 250 V. | For a 40 W sine wave into 6 Ω, the amplifier output is about 15.5 V RMS or 21.9 V peak. The voltage across C2 depends on frequency and the reactive network; select its voltage rating with margin, not solely from amplifier output voltage. |
2. Frequency response and ripple (SPL)
In this section we will see the shape of the model's frequency response curve. Before entering the numbers, it is worth setting the band in which the model–measurement comparison makes sense: approximately 300 Hz–10 kHz. It is the band in which LspCAD issues a prediction and in which the planned measurement procedure can, in principle, compare. We do not call it "validable" because that would presuppose that the experimental method allows validation with sufficient uncertainty; we only state that it is the band where the comparison makes sense to attempt.
Within that band, the response depends on four contributions that should be distinguished: the drivers' own response, the geometry and adopted offsets, the action of the crossover, and possible reflections from the measurement environment. If a discrepancy appears, it should not automatically be attributed to only one of the four; that is what causal diagnosis is for (section 7).
Below ~300 Hz, on the other hand, things change. In that zone the response depends on factors that LspCAD does not model in this simulation: real baffle step, cabinet dimensions and losses, woofer alignment, presence or absence of a port, room position and room gain. A discrepancy measured below 300 Hz is not sufficient on its own either to blame or to exonerate the crossover: the woofer series inductor also affects this region. It is a different question, and it deserves a separate analysis.
Between 200 Hz and 10 kHz, the model's on-axis response stays in an approximate range of 88–91 dB. There are no deep holes at the two main transitions: woofer→midwoofer, model acoustic crossover towards 700–900 Hz; midwoofer→tweeter, towards 4–5 kHz. Above 10 kHz there is a progressive fall: ~88–90 dB towards 10 kHz, ~85 dB towards 15 kHz, ~78–83 dB towards 20 kHz.
Estimated ripple 400 Hz – 10 kHz (model)
| Band | Estimated ripple |
|---|---|
| 400 Hz – 4 kHz | ±0.7 to ±1 dB |
| 4 kHz – 10 kHz | ±1.5 to ±2 dB |
The fall at extreme treble is a characteristic of the model. It will need to be verified by measurement and listening with the real cabinet, because it may be due both to the tweeter's response and to the action of the filter or the geometry. PENDING MEASUREMENT
3. Impedance and amplifier compatibility
In this section we will see what load the speaker presents to the amplifier. Impedance is not a fixed value: it changes with frequency. What interests us is knowing how low it drops in the worst case and what phase it has at that point, because that determines how demanding the cabinet is for the amplifier.
| Parameter | Value | Evidence |
|---|---|---|
| Impedance minimum (model) | Provisional, close to 4.9 Ω at ~1.4 kHz | PREDICTION PENDING MEASUREMENT |
| Phase at that point | Moderate; the model does not show a pronounced reactive extreme in the minimum zone | PREDICTION |
| Operating band | The impedance modulus stays approximately in the 5–8 Ω zone over most of the modelled band | VISUAL ESTIMATE FROM GRAPH |
| Nominal impedance | ≈6 Ω | PREDICTION PENDING MEASUREMENT |
| Model EPDR | >≈3.5 Ω, estimate dependent on the predicted curve | ESTIMATED PENDING MEASUREMENT |
What does the model tell us? That the cabinet does not present the combination of very low impedance and high phase that can increase dissipation in the output stage. EPDR estimates that demand. The change in frequency response depends on the voltage divider formed by the loudspeaker’s complex impedance and the amplifier’s output impedance at each frequency. The model suggests an EPDR above ≈3.5 Ω, but that value must be verified from the measured impedance and phase of the finished cabinet. The specific compatibility with a valve amplifier must be checked with the real impedance of the cabinet and the amplifier's output impedance. INFERENCE PENDING MEASUREMENT
On the amplifier output tap. It is not generally established that 4 Ω or 8 Ω are "the correct one". The tap recommended by each amplifier's manufacturer for this nominal impedance (~6 Ω) should be used, documented and kept constant between comparisons. With valves it may be informative to measure both taps and document how the frequency response changes due to the amplifier's output impedance. INFERENCE
4. System geometry and directivity
Physical geometry of the acoustic centres
In this section we will see how each driver is placed on the baffle and what acoustic consequences that placement has. The speaker baffle is flat and vertical (90° to the floor). The three transducers are mounted flush, without a step or tilt. The system reference point is the tweeter-midwoofer junction: that is the design axis, and it will also be the listening axis.
Each transducer has an effective acoustic centre depth that has been adopted in the model. This depth is related to the driver geometry, its motor and its acoustic behaviour. It should not be interpreted as an invariant physical coordinate of the transducer: the acoustic centre can vary with frequency, and it is one of the main uncertainties of the model. ADOPTED IN THE MODEL
| Driver | Vertical dy | Acoustic offset adopted in the model | Note |
|---|---|---|---|
| Tweeter | +38.5 mm | 0 mm (reference) | System reference offset |
| Midwoofer | −38.5 mm | 5 mm behind the baffle | Adopted for the model |
| Woofer | −223.5 mm | 20 mm behind the baffle | Adopted for the model |
What does this imply for construction? That no step or baffle tilt is needed. The acoustic offsets adopted by LspCAD are included in the model and are absorbed by the crossover design. Whether those offsets match the real ones of the physical units is a question to be verified by measurement. PHYSICAL MOUNTING: CONFIRMED · ACOUSTIC OFFSETS: ADOPTED IN MODEL, PENDING MEASUREMENT
Why do these millimetres matter? At 4.5 kHz —the crossover between midwoofer and tweeter— the wavelength is about 76 mm. The 5 mm difference between the offsets adopted for midwoofer and tweeter is equivalent to 6.6 % of a wavelength, about 24° of phase. At 750 Hz, the wavelength is about 457 mm; the 15 mm difference between the adopted offsets is equivalent to 3.3 % of a wavelength, about 12° of phase. Do not confuse these 15 mm of offset difference with the 185 mm of geometric separation between centres, which are the ones used for the notch geometry. CALCULATED
Angles to the listening axis (at 3 m)
Let's now see what angle each driver forms with the listener's ear. With the ear on the axis (the tweeter-midwoofer junction) and at 3 m distance, the vertical angles to each acoustic centre are:
| Driver | Vertical angle | Status |
|---|---|---|
| Tweeter | arctan(38.5/3000) ≈ +0.73° | On axis |
| Midwoofer | arctan(−38.5/3000) ≈ −0.73° | On axis |
| Woofer | arctan(−223.5/3000) ≈ −4.26° | Within tolerance |
The conclusion is reassuring: the critical mid-tweeter crossover (~4–5 kHz) occurs practically on axis, because the two drivers involved are only ±0.73° from the listening point. The woofer, at 4.3° below the axis, remains in the model within a small angle with respect to its crossover with the midwoofer. CALCULATED
Horizontal directivity (model)
Directivity describes how sound is distributed in space. In the horizontal plane, the model predicts an accumulated loss of ~2 dB at ±20° and ~4–6 dB at ±30° in the 6–10 kHz band, with progressive narrowing above. PREDICTION
| Horizontal angle | Loss in 6–10 kHz |
|---|---|
| 0° (on axis) | 0 dB |
| ±10° | < 1 dB |
| ±20° | ~2 dB |
| ±30° | ~4–6 dB |
| ±60° | Very high attenuation |
Vertical directivity — the main compromise
In the vertical plane, the model shows two cancellation notches, one for each crossover between ways. A notch is a deep, narrow level drop that appears at a specific direction and frequency, due to destructive interference between two drivers.
Midwoofer-tweeter crossover notch (~4.5 kHz)
It is the critical notch, because it is closer to the vertical listening axis. Its relative drop is ≈18–20 dB with respect to the on-axis level, and it appears situated at ±25–30° measured from the tweeter-midwoofer junction.
Conditional ideal geometric estimate. The adopted vertical geometric separation between tweeter and midwoofer is 77 mm. At 4.5 kHz the wavelength is ~76 mm. The total cancellation condition (d·sin θ = λ/2) gives sin θ = 38/77 ≈ 0.493 → θ ≈ 29.5°. At 3 m distance, ±29.5° is equivalent to ±1.7 m of vertical displacement. CALCULATED
Position observed in the model. LspCAD reproduces that notch at ±25–30° with the declared drop of 18–20 dB. LspCAD PREDICTION
Woofer-midwoofer crossover notch (~1 kHz)
There is a second cancellation minimum in the ~1 kHz zone, corresponding to the crossover between woofer and midwoofer. With 185 mm separation between acoustic centres, the ideal geometric approximation (d·sin θ = λ/2), valid only for compatible phases and amplitudes, places the cancellation at ±60–70° vertical around 1 kHz. CALCULATED
Numerical verification: for θ = 65°, sin θ = 0.906 → λ/2 = 185 × 0.906 = 168 mm → λ = 335 mm → f ≈ 1,024 Hz. For θ = 70°, sin θ = 0.940 → λ/2 = 174 mm → λ = 348 mm → f ≈ 986 Hz. The notch therefore lies between 986 Hz and 1,024 Hz depending on the exact angle.
At 3 m distance, ±65° vertical is equivalent to > 6 m of vertical displacement with respect to the axis. It is completely outside any reasonable listening position. It does not correspond to direct listening from a typical seat, but it may contribute to the room’s reflected field; its significance should be checked with angular measurements.
Vertical variation according to the model at 3 m
| Vertical misalignment with respect to the axis | Resulting angle | Behaviour in the model |
|---|---|---|
| 0 mm | 0° | Optimal |
| ±100 mm | ±1.9° | Small variation region |
| ±250 mm | ±4.8° | Small variation region; matches the recommendation of section 6 |
| ±500 mm | ±9.5° | Changes in response begin to be visible from here |
| ±700 mm | ±13° | Visible degradation at 4–5 kHz |
| ±1,090 mm | ±20° | Approach to the region where the depression associated with the mid-tweeter notch begins to become significant |
| ±1,600 mm | ±28° | Full mid-tweeter notch |
| > 6 m | ±65–70° | Woofer-mid notch (theoretical only, outside any real position) |
The correct formulation is this: the model shows a region of relatively small variation around the design axis; at 3 m it covers approximately ±25 cm (±4.8°). Degradation becomes progressively more visible as one moves away from the axis. It should not be stated "without audible effect" because the real response of the cabinet, the real width of the notch and the room acoustics are not yet characterised. CALCULATED PREDICTION PENDING MEASUREMENT
On the green patch at 1 kHz off-axis. In the model's vertical map a greenish zone appears in the 1 kHz band at intermediate angles. It should not be interpreted as cone breakup (an abrupt deformation of the diaphragm). The simplest cause in this model is the residual contribution of the woofer above its crossover, favoured by the first-order electrical slope of L3. INFERENCE
Joint reading of Figures 4 to 8
The five figures cover the same phenomenon from two complementary perspectives. Figures 4, 5 and 6 are discrete cuts: Figures 4 and 5 show the response at specific angles, and Figure 6 shows it at specific frequencies (1, 2, 4, 6, 8 and 10 kHz). Figures 7 and 8 are continuous views: colour maps on the frequency-angle plane.
- Figure 6 (polar diagram): shape of the modelled lobe at each specific frequency.
- Figure 7 (horizontal map): summarises the entire horizontal behaviour of the model in a single image.
- Figure 8 (vertical map): the most critical of the three. It makes visible the two notches that the text describes.
Taken together, the five figures are coherent with what the report maintains by calculation. They do not "confirm" the real cabinet: they confirm the behaviour of the model built with those .frd/.zma. LspCAD PREDICTION CALCULATED PENDING MEASUREMENT
5. Sensitivity and required power (under explicit hypotheses)
SYSTEM SENSITIVITY: ND
89 dB/W/m — CALCULATION HYPOTHESIS, NOT AN LspCAD OUTPUT
Before making any calculation, it must be made clear: the real sensitivity of the finished cabinet will not be known until it is measured. The value of 89 dB/W/m is a working hypothesis, not a property of the cabinet. The calculations that follow must be read as conditional on that hypothesis and on a free-field scenario. PENDING MEASUREMENT
Three levels of treatment of a numerical value
To avoid confusing what is a prediction with what is a specification, this document distinguishes three levels for any numerical value of the system (89 dB, impedance, EPDR, 95 dB SPL, 200 Hz–10 kHz response):
- Level 1 — Prediction. "The working hypothesis is approximately 89 dB/W/m."
- Level 2 — Validation. "The measurement gives X; the comparison with the prediction is carried out according to the pre-registered criterion."
- Level 3 — Specification. "The cabinet is characterised as X dB/W/m."
You do not go directly from level 1 to level 3. The distance between the two is the measurement, and that distance cannot be skipped without losing traceability.
Modelled scenario. Suppose we want 80 dB SPL continuous plus 15 dB crest, that is, 95 dB peak, at 3 m distance and in free field. To maintain a conservative criterion, the following calculations consider the SPL produced by a single cabinet. For this calculation, an energetic sum of +3 dB between two cabinets is assumed. It is not a universal gain: it depends on channel correlation, frequency, listening position and the acoustic field.
In the modelled scenario, maintaining 80 dB SPL continuous plus 15 dB crest at 3 m requires approximately 1.1 W average power and a peak capability of ~36 W in one cabinet (or ~18 W/channel in stereo with the ideal sum). These figures are a calculation conditional on the sensitivity hypothesis and the free-field scenario, not a specification of the cabinet.
Note on baffle step. LspCAD with a piston model does not include the effect of the real cabinet's baffle step. Baffle step is a gradual transition between full-space and half-space radiation. The ideal level difference can approach 6 dB, but the actual transition depends on the baffle, drivers and environment. In the band below ~300 Hz, the response also depends on cabinet dimensions and losses, woofer alignment, presence or absence of a port, room position and room gain. A measured discrepancy in this band should not be used to falsify the crossover. PENDING MEASUREMENT DO NOT DECIDE YET
SPL / required power calculator (conditional on the working hypothesis)
You can use the calculator to explore how the required power changes when modifying sensitivity, distance or target SPL. The formulas are: SPL_1W@d = sens − 20·log10(d) and P = 10^((target SPL − SPL_1W@d − factor)/10). The results are per channel. To reproduce the calculations of section 5, leave sensitivity at 89 dB/W/m.
Important caution. The above numbers are a calculation conditional on seven assumptions that are not yet verified: (1) real sensitivity, (2) driver thermal behaviour, (3) driver THD at those levels, (4) compression, (5) real listening distance, (6) room gain, (7) crest level actually demanded. They should not be presented as a demonstrated property of the cabinet. INFERENCE PENDING MEASUREMENT
Additional precaution. If one amplifier enters clipping while another retains margin, the difference is one of available dynamic capability, not proof that one has a certain superior timbral quality. INFERENCE
6. Listening positioning
Recommended configuration
This section is the most practical in the report. Here the model's predictions are translated into a concrete recommendation on where to sit and how to place the speakers. The recommendations are based on the model's directivity and response; verification in the real room remains pending.
| Parameter | Optimal value | Calculation basis |
|---|---|---|
| Ear height | Tweeter-midwoofer junction | System design axis |
| Listening distance | 2.5 – 3.5 m | Model directivity and SPL |
| Horizontal angle per side | 20–25° | Off-axis ~2 dB up to 20° in the model |
| Tweeter separation (at 3 m) | 2.2 – 2.8 m → 20–25° per side. Central working zone: 2.4 – 2.6 m → approx. 22–23.5° per side. | 2 × 3 × tan(θ) |
| Toe-in | Slight, aiming at a point between the head and 30 cm behind | Off-axis symmetry |
| Preferred vertical listening region | Around the design axis; at 3 m, ±25 cm corresponds to ±4.8° | Within the model's small variation region |
Coherent vertical scale (at 3 m distance). To give you an idea of how much margin there is:
- ±25 cm → model small variation region.
- 40–50 cm → still relatively close to the axis, but can already produce changes in the response around 4–5 kHz.
- ±70 cm → clearly visible variation.
- ±1.1 m → approach to ±20°, the region where the depression associated with the mid-tweeter notch begins to become significant.
What to avoid
- Do not lower the cabinet "so that the tweeter points at the ear". The design axis is the tweeter-midwoofer junction, not the centre of the tweeter. If the tweeter points at the ear but the midwoofer is below, the crossover between the two shifts off-axis.
- Do not separate the speakers to the classic equilateral triangle (30°). The model predicts 4–6 dB less treble energy at that angle, which will translate into a duller scene.
- Do not tilt the cabinet forwards or backwards. The baffle is designed vertical; tilting it changes the relative vertical angles between drivers and shifts the cancellation notch.
- Do not obstruct the tweeter-ear path with armrests, reflective low tables or high backs between speakers.
- Keep vertical listening reasonably close to the design axis. Displacements of 40–50 cm correspond approximately to 7.6–9.5° at 3 m and can modify the response around 4–5 kHz. PENDING MEASUREMENT
Practical verification
Concrete set-up, step by step:
- Measure the height of the tweeter-midwoofer junction on the finished cabinet.
- Measure your ear height seated at the listening point.
- Adjust chair or stand until they coincide.
- With 3 m distance, separate the tweeters 2.2–2.8 m (20–25° per side); 2.4–2.6 m correspond to 22–23.5° as the central working zone.
- Apply a slight toe-in until the axis of each cabinet points to a point between your head and 30 cm behind.
- Check that there are no obstacles between tweeters and ears.
7. Evidence classification and validation protocol
This section serves two functions. First, it classifies all the statements in the report according to their level of evidence. Second, it describes the specific protocol that will need to be followed to convert predictions into measurements. It is the most important section of the document from a methodological point of view.
CONFIRMED
The values of the five components, the parallel topology (C1 + R2 in series, L4 in shunt), the physical mounting geometry (the vertical positions and geometric centres of the three drivers on the baffle) and the acoustic offsets adopted in the model. Also that certain .frd/.zma files have been used as input data.
PREDICTION / SIMULATION
Everything that LspCAD produces from those files: transfer functions, on-axis response, resulting impedance, phase and modelled directivity. The polar maps (Figures 6, 7 and 8) are the graphical representation of the predicted directivity, not a confirmation of the physical cabinet.
CALCULATED
The physical separation between the geometric centres of the drivers, the geometric angles at 3 m, the vertical displacements, the conditional geometric estimates of the two notches, the required power under explicit hypotheses (working sensitivity of 89 dB/W/m, free field, 3 m) and all associated arithmetic.
INFERENCE
The choice of 20–25° per side as the preferred listening zone; the listening height at the tweeter-midwoofer junction; the advisability of a slight toe-in; the practical consequences derived from the model. Also the interpretation of the 4–5 kHz residue, the reading of the off-axis patch at 1 kHz and the characterisation of the expected behaviour of the system with valve stages.
PENDING MEASUREMENT
Real sensitivity, exact Zmin, nominal impedance of the finished cabinet, exact phase, THD at 80–95 dB, thermal compression, maximum clean SPL, real directivity, real in-room response and real baffle step. In particular:
- The crossover simulation does not prove that the midwoofer reproduces 95 dB peak with low distortion: neither excursion, nor THD, nor compression are modelled.
- The real baffle step of the cabinet can introduce an imbalance below 300 Hz that the simulation does not reflect.
- The coincidence between the acoustic offsets adopted in the model and the effective offsets of the physical units must be verified by measurement.
- The model EPDR (>≈3.5 Ω) must be verified from the measured impedance and phase of the finished cabinet.
DO NOT DECIDE YET
Changes whose meaning depends on data that do not yet exist. Any decision on these points, taken before measuring, is speculation:
- Tweeter R1 (attenuation or level adjustment).
- Change of L3 (for example to 2.0–2.2 mH).
- Change of C1 (reduction towards 27–22 µF or another variation).
- Modification of R2.
- Any component presented as a sonic improvement without verification.
- Driver substitution.
Frozen reference model — MR-A-01
Crossover A v2.4 is frozen as Reference Model MR-A-01. The freezing is not limited to the five crossover values: it covers all elements that, if modified, invalidate the comparison between prediction and subsequent measurement.
Frozen elements:
- LspCAD version and exact build.
- .frd files used: name, date, origin and SHA-256 checksum of each.
- .zma files used: name, date, origin and SHA-256 checksum of each.
- Assumed geometry: positions and separations between geometric centres.
- Acoustic offsets adopted: the three values, with adoption date.
- Crossover components: nominal values and real DCR introduced in the model.
- Simulation conditions: range, resolution and window type if applicable.
- Date and author of model closure.
If at any later time any of these elements must be modified, MR-A-01 is not edited. MR-A-02 is created and what changed and why is documented. The comparison with measurements is always made against the model that generated the prediction, not against a later version that already knows the result.
Interaction with the amplifier — separate later experiment
The amplifier Zout sweep is not part of the crossover validation. It is a later experiment, with its own design. First: cabinet → crossover → drivers → real response → real impedance. Then: amplifier Zout → interaction with Zspk(f). The two questions are kept separate.
Objective: quantify ΔSPL(f) produced by the Zout,amp × Zspk(f) interaction. Sweeps are performed with Zout = 0.1 / 0.5 / 1 / 2–3 Ω, keeping everything else constant. It is not seeking "better sound"; it is seeking to quantify the interaction. Coherent with the project's valve objective.
Interpretation criterion and diagnosis
When the physical measurement arrives, the comparison between prediction and measurement is evaluated on two orthogonal axes. They are not mixed.
Axis 1 — Comparison status
- CONFIRMED / COMPATIBLE
- DISCREPANCY
- INCOMPATIBLE / CANDIDATE FOR FALSIFICATION
- NOT DECIDABLE
Axis 2 — Diagnosis
- SUFFICIENT MODEL — there was an applicable prediction and the result confirms it.
- INSUFFICIENT MODEL — the model does not contain the evaluated phenomenon. Applicable to THD, compression and, in part, baffle step.
- EXCESSIVE UNCERTAINTY — the procedure does not allow discrimination.
- CAUSE PENDING — there is a discrepancy but the cause is not determined.
Four causes of discrepancy
- Definition error. The prediction was not formulated with sufficient precision to be compared.
- Input error. The prediction was well defined, but the model inputs do not represent the physical system.
- Model error. Correct inputs, phenomenon not contained.
- Experimental error. The measurement does not have sufficient resolution or is contaminated by the procedure.
The first question before any discrepancy is not "did the model fail?", but "was the prediction I am comparing against comparable?". The second is "what is the difference attributed to?".
Note on FALSIFIED. The word FALSIFIED is reserved for the case in which six conditions are simultaneously verified: the prediction was correctly defined, the model contains the evaluated phenomenon, the model inputs are correct, the experimental procedure is valid, the uncertainty is correctly estimated, and the discrepancy persists. In this project it will probably not be used; the operative vocabulary is CONFIRMED / DISCREPANCY / INCOMPATIBLE / NOT DECIDABLE, with causal diagnosis as accompaniment.
Validation protocol
Revised sequence. The order matters: first the real cabinet is measured, and only then is the crossover adjusted.
- Measure the impedance of each driver installed in its enclosure (woofer, midwoofer, tweeter). This is the base measurement, not the isolated T/S parameters.
- Measure the individual response of each way with its real physical geometry and current impedance. If considered necessary, additional characterisation of the driver parameters.
- Measure the impedance of the original cabinet, if the old crossover still exists, to have a comparative reference.
- Build exactly Crossover A (five components, C1 + R2 in series, L4 in shunt) with the flat vertical baffle.
- Measure the new cabinet with Crossover A in the same measurement configuration as the original, so that they can be compared: original → Crossover A → subsequent modifications.
- Determine the effective baffle step by combining a near-field measurement of the woofer and a far-field measurement with gating, joining both curves using the appropriate measurement procedure. The direct difference between a measurement at 10 cm and another at 1 m must not be interpreted as the baffle step: distance, acoustic geometry, relative contribution of each transducer, near-field/far-field conditions and interaction with the enclosure all change.
- Measure on-axis response at 1 m (with gating or appropriate technique), with the microphone at the tweeter-midwoofer junction.
- Measure THD of the 7.7 cm midwoofer at 90 dB and at 95 dB SPL measured at the reference position, between 800 Hz and 4 kHz. With L4 in shunt, the midwoofer has a second-order high-pass electrical network and should not excursion as much as with a first order. If THD exceeds 3 % at 95 dB SPL measured at the reference position, reducing C1 from 33 µF towards 27–22 µF can be considered and the measurements repeated. The decision should not be taken on THD alone: the acoustic sum woofer–midwoofer, response and phase in the 500 Hz–1.5 kHz zone must be verified simultaneously, because the objective is not only to protect the driver but to preserve the system integration.
- Measure horizontal directivity (0°, ±10°, ±20°, ±30°, ±45°, ±60°) to compare with the polar map of Figure 7.
- Measure vertical directivity (0°, ±5°, ±10°, ±20°, ±30°) to compare with the polar map of Figure 8 and check the position of the two notches.
- Measure global THD at 80 dB, 90 dB and 95 dB SPL at the reference position.
- Measure at 3 m at the real listening position, in the room of use.
- Only then decide whether the crossover needs additional modification. Any change of L3, C1, R2 or R1 must be simulated with the real DCR of the new component and verified in response, impedance and woofer–midwoofer sum before accepting it.
On tweeter protection during the first tests: known and stable amplifier, check for absence of DC, initially minimum volume, first measurement at low level, progressive increase, THD/SPL monitoring, and avoid any power test before knowing the real response. INFERENCE
On future driver changes
Crossover A should not be designed already thinking about replacing the midwoofer and tweeter. The correct sequence is:
- Phase 1 — original drivers: 27 cm paper+mica + 7.7 cm paper + 5 cm paper → Crossover A → measure → listen.
- Phase 2 — modernisation (if decided): if midwoofer and tweeter are replaced with modern equivalents, the crossover will have to be redone from the new measurements. It would not be valid to automatically keep 33 µF, 0.82 mH, 2.2 Ω, 2.70 µF: Re, Le, Fs, impedance, sensitivity, response and directivity would change.
The possible later transformation to a FAST-type system (27 cm + ~4" paper full-range) is a different line of research, not a modification of Crossover A. It should not be mixed into this document.
8. Conclusion
Crossover A is an experimental reference mature enough to move from simulation to construction. It is not yet a crossover validated on the physical units.
Throughout the report we have seen that the simulation predicts a reasonably uniform response in the main band (±0.7 to ±1 dB up to 4 kHz; ±1.5 to ±2 dB from 4 to 10 kHz), with the woofer-midwoofer transition placed low enough to reduce the woofer's contribution in the region where its directivity begins to narrow. The integration of the tweeter with a first-order filter (C2 = 2.70 µF) and no deliberate series attenuation resistor is plausible. The midwoofer, with a second-order high-pass electrical network (C1 + R2 in series, L4 in shunt), uses its natural rolloff as the upper band limit, without an additional electrical low-pass.
The geometry also fits: the baffle is flat and vertical, with the three drivers flush, and the acoustic offsets adopted (+0 mm tweeter, −5 mm midwoofer, −20 mm woofer with respect to the baffle) are included in the simulation. The model's horizontal directivity shows ~2 dB loss at ±20° and ~4–6 dB at ±30° in 6–10 kHz, consistent with the polar maps. Vertical directivity presents a small variation around the axis up to approximately ±25 cm at 3 m (±4.8°), with two cancellation notches consistent with physics: mid-tweeter at 4.5 kHz at ±25–30°, and woofer-mid at ~1 kHz at ±60–70° (the latter completely outside any real listening position).
The model impedance presents a provisional minimum close to 4.9 Ω with moderate phase, without the simultaneous combination of very low impedance and high phase. The estimated EPDR (>≈3.5 Ω) must be verified from the measured impedance and phase of the finished cabinet. Under the modelled scenario of 80 dB continuous plus 15 dB crest at 3 m, and with the working sensitivity of ~89 dB/W/m, maintaining that SPL requires approximately 1.1 W average power and a peak capability of ~36 W in one cabinet (or ~18 W/channel in stereo with the ideal sum). These are conditional figures, not demonstrated properties.
The relevant question, at this point, is no longer a fine adjustment of C2 on the model. It is another: what is the real sensitivity? The real impedance? How does the cabinet respond at 3 m in the room? How much THD does it produce at 90–95 dB SPL? Where is the real baffle step? Until those answers are obtained, Crossover A must be treated as a working reference, not as an acoustically validated design. And precisely for that reason it is worth building and measuring.
The next step is no longer to keep optimising the simulation: it is to build and measure MR-A-01.
APPENDICES
Supporting material. Appendix A develops the evidence label system specific to this report. Appendix B gathers the sources and traceability.
APPENDIX A — HOW TO READ THE EVIDENCE LABELS
Appendix A — How to read the evidence labels
Each statement in the report carries an evidence label indicating its level of support. This appendix develops the summary legend from the side aside (visible on desktop) for the mobile and print versions, where the aside is hidden.
The seven labels, one by one
| Label | What it means | What would be needed to upgrade it |
|---|---|---|
| CONFIRMED | Schematic data, topology, mounting or verifiable starting value. | It is already at the highest level. It is only degraded if a documentary contradiction or a physical measurement disproves it. |
| PREDICTION / SIMULATION | Result of the LspCAD 6.20 pro engine applied to the starting .frd/.zma files. | Physical measurement of the finished cabinet under the same field condition. |
| VISUAL ESTIMATE | Approximate reading of a simulated graph, without digitisation or numerical processing. | Digitise the curve or, better, measure the phenomenon on the real cabinet. |
| CALCULATED | Arithmetic derivation with explicit formula and known starting data. | Experimentally verify the result with an instrument. |
| INFERENCE | Technical or application interpretation, not an objective property of the system. | Measurement or documentary confirmation that turns the interpretation into fact. |
| PENDING MEASUREMENT | Requires experimental validation on the physical cabinet. It is not that information is missing: it is that the object to be measured does not yet exist. | Build the cabinet, measure and compare with the recorded prediction. |
| DO NOT DECIDE YET | Change whose meaning depends on data that does not yet exist. Any decision here without measuring is speculation. | That the data conditioning the decision exists. |
| IMG-XX | Photographic figure reference. It is not an evidence label in the strict sense, but a visual locator. | — |
How several labels combine in the same sentence
Sometimes the same sentence carries two different labels, because different parts of the sentence have different levels of evidence. For example:
The same sentence contains two levels: what the model predicts (4.9 Ω) and what the real cabinet will say when measured (does not yet exist). The reader who reads the two labels as if they said the same thing will miss the real state of the question.
APPENDIX B — SOURCES AND TRACEABILITY
Appendix B — Sources and traceability
Technical sources
- LspCAD 6.20 pro — Speaker simulation software. All the frequency response, impedance, transfer function and polar map figures in this report come from this tool applied to the starting .frd/.zma files of the three transducers. (accessed 2026)
- Meta, ChatGPT, DeepSeek and Claude — Assisted writing of the report, formulas, geometric calculations and design of the validation protocol. The authorship of the document is shared with maty @nauscopio. (2026)
- Datasheets from the manufacturers of the 27 cm woofer, 7.7 cm midwoofer and 5 cm tweeter — Nominal specifications (Re, Le, Fs, sensitivity) used as initial context, not as inputs to the LspCAD model.
Image credits
- Figures 0 to 8: LspCAD 6.20 pro simulation captures and report screens, prepared for this document.
- Crossover A schematic (Figure 0): design drawing generated from the frozen component list.
Version and freezing
- Crossover A · V2.4 · Reference Model MR-A-01 frozen before cabinet construction. Any modification of the frozen elements creates MR-A-02.
- Report date: October 2026. Language: English.
About how this report was prepared
This report was prepared with cross-auditing by several AI tools (Meta, ChatGPT, DeepSeek, Claude, Mistral) and manual verification of primary sources. The tools were used for cross-checking and error detection; the final statements are the author's responsibility. The evidence methodology is described in Appendix A.