Solar Oracle Walkman - Decomposed Authorship and Distributed Green Certificate as Generative Radio Art (The technical document)

Solar Oracle Walkman

Solar Oracle Walkman is an ongoing artistic research and engineering project that explores whether the electrical behaviour of a handmade dye-sensitized solar cell (DSSC) can act as a material root for identity, evidence and sound.

The project adapts the I-V curve tester developed by Marc Dusseiller into a physical interface for measuring a DSSC through sixteen resistor loads. A button press captures one I-V curve, sends it to a local identity service and, after a successful verification, activates a six-channel World Radio Mixtable. The mix is calculated from the measured photovoltaic features rather than selected arbitrarily.

At its present stage, Solar Oracle Walkman is a minimum viable prototype for making a green credential audible. It is not yet a certified PUF system, a deployed renewable-energy certificate network or a trained RAVE synthesizer. The long-term goal is infrastructure for distributed green credentials in which a physical energy source can establish a traceable identity and authorize a reproducible generative-audio process.

Project status

Layer	Present status
DSSC I-V measurement	Implemented with a sixteen-resistor ESP32 tester.
Physical button workflow	Implemented. One press requests one sweep and sends one result to the receiver.
Local display	Implemented on the tester's e-paper display for the local I-V curve and electrical values.
DSSC identity	Experimental two-cell prototype using Conv1D classification, a 7D physical fingerprint and open-set rejection.
Local registration	Two DSSCs are enrolled locally as N719-G and N719-B.
Local oracle records	Implemented as hash-linked event, dataset, memory and replay records.
World Radio Mixtable	Implemented as a deterministic six-channel internet-radio mix.
Evidence dashboard	Implemented as a live browser interface containing the curve, identity evidence and radio mapping.
Smart-contract prototype	Implemented and tested as an optional signed-report authorization experiment, but not active in the current installation.
RAVE	Dataset and control scaffold only. No trained RAVE model is used in the current MVP.
Distributed green credential	Long-term research goal.

Concept: a material root for a DSSC PUF

A handmade DSSC contains variations in coating, porosity, dye uptake, interfaces, sealing, contact resistance and ageing. Solar Oracle Walkman treats the resulting I-V curve as a possible material signature. The measurement is not only an energy reading; it is investigated as evidence that a specific physical solar cell is present under a specific set of conditions.

The system does not claim that two cells are enough to prove a secure physical unclonable function. The current work is an operational hypothesis and an artistic prototype. A formal PUF claim would require a much larger population, calibrated illumination, cross-day and cross-instrument tests, and measured false-accept and false-reject rates.

The material-root workflow is:

light
 -> handmade DSSC
 -> sixteen resistor loads
 -> I-V curve
 -> normalized physical fingerprint
 -> Conv1D identity decision
 -> accepted / unknown
 -> local oracle event
 -> World Radio Mixtable
 -> audible evidence

Hardware

Sixteen-resistor I-V tester

The measurement instrument is based on Marc Dusseiller's open I-V curve tester. The Solar Oracle adaptation uses an ESP32-class controller, a resistor multiplexer, a physical button, an addressable status LED and a tricolour e-paper display.

The tester measures sixteen operating points. For every point it records voltage, current and power. The sweep is used to estimate the open-circuit voltage, short-circuit current, maximum-power point and fill factor, while the complete sequence preserves the shape of the curve.

The public tester repository is listed in the Source section below.

Button-triggered measurement

The installation does not continuously classify the cell. A physical button edge starts a new measurement. This gives every identity event a clear physical beginning and prevents a stale curve from silently creating a new audio event.

One press performs the following actions:

1. Reset the resistor channel.
2. Measure all sixteen loads.
3. Store voltage, current and power arrays.
4. Send the complete measurement to the Mac receiver over HTTP.
5. Refresh the local e-paper I-V display.
6. Show success or rejection through the tester status light.

The network request begins before the slower full e-paper refresh, allowing identity analysis and display refresh to proceed in parallel.

E-paper display

The tester display shows the Solar Oracle Walkman title, the local I-V plot and electrical values such as maximum power, approximate Isc, approximate Voc and fill factor. The built-in tricolour panel uses a full refresh.

The Mac also produces a compact e-paper status representation containing the decision, authorization, confidence, repeatability and agreement. A dedicated display backend can consume this representation when another e-paper device is used.

Measurement applications and methods

Continuous collector

The continuous collector periodically reads the tester's HTTP data endpoint. It is used for enrollment sessions, repeated measurements and controlled experiments. It checks that all sixteen points are present, rejects insufficient signal and records both normalized CSV data and untouched raw payloads.

Button-aware collector

The button-aware collector waits until the press counter or curve data changes. It is useful when repeated measurements must remain associated with separate physical actions.

Passive push receiver

The principal installation mode is a local receiver on the Mac. The tester sends a measurement directly to its trigger endpoint. The receiver then:

1. Stops the previous radio mix.
2. Creates a timestamped measurement session.
3. Stores the raw payload and standardized curve data.
4. Runs Conv1D and physical-fingerprint verification.
5. Creates an accepted or rejected oracle event.
6. Builds the World Radio Mixtable after acceptance.
7. Regenerates the evidence dashboard once the complete event is ready.
8. Starts six local audio players or preserves silence after rejection.
9. Returns the authorization result to the tester.

Diagnostic watcher

A lightweight diagnostic watcher records the tester endpoint for a short period. It is used to inspect button behaviour, network timing and changing measurements; it is not the normal performance interface.

Enrollment and registration

Physical enrollment

Enrollment is based on repeated I-V curves collected from a known DSSC. Each session stores:

• the measurement identifier;
• sixteen indexed voltage/current points;
• the raw tester response;
• collection time and measurement context;
• hashes used for provenance;
• the declared DSSC label.

The current prototype contains two locally enrolled cells:

Local identity	Stable motif	Primary sound address
N719-G	Japan / ambient	Channel 1
N719-B	France / jazz	Channel 2

These cultural and musical addresses are compositional mappings. They are not claims about the geographic origin of the physical cells.

Local registration boundary

"Registered" currently means that the cell label, its trained identity model and its stable radio motif are recognized by the local Solar Oracle installation. There is not yet a public registration form, owner-wallet binding, certificate authority or active on-chain DSSC registry.

Optional smart-contract experiment

An optional EIP-712 smart-contract prototype can verify a signed inference report, a model hash, a threshold, a timestamp and replay-protection values. The contract does not execute Conv1D and does not itself register a DSSC. The current World Radio performance loop uses local oracle records rather than an active blockchain transaction.

Conv1D identity and physical fingerprints

Normalized curve input

The curve is normalized by Voc and Isc and resampled to 128 points. This reduces dependence on absolute light intensity and cell area while preserving relative curve shape. It does not remove all lighting, contact, ageing or instrument effects.

Two-board Conv1D classifier

The present live model distinguishes the two enrolled classes N719-G and N719-B. It is a closed two-class predictor guarded by an open-set physical-distance check. A curve is not accepted merely because one class is closer than the other.

The live decision requires:

• consistent Conv1D predictions;
• consistent physical-fingerprint predictions;
• agreement between the two methods;
• a fingerprint distance inside the selected cell's calibrated domain.

If these conditions fail, the result is unknown. Unknown measurements do not enter the accepted musical memory and do not start another cell's sound.

Current 7D physical fingerprint

The current explicit physical fingerprint contains:

1. Fill factor (FF)
2. Vmpp / Voc
3. Impp / Isc
4. Normalized series-resistance proxy (Rs*)
5. Normalized shunt-resistance proxy (Rsh*)
6. Sum of normalized curve curvature
7. Normalized area under the I-V curve

Eight-dimensional latent music control

A separate one-cell Conv1D autoencoder experiment produces an eight-value learned latent vector. That vector can be mapped to tempo, voltage character, current character, fill factor, knee, sag, leakage and entropy controls.

This distinction is important: the current two-cell identity gate uses the explicit 7D physical fingerprint, while the 8D latent fingerprint/control layer is the intended bridge toward a future RAVE model. The future system is expected to condition neural audio generation with the accepted DSSC identity and this learned latent state.

Local oracle, evidence and memory

Every accepted or rejected decision creates a local oracle event. The event links the measured curve, identity result, confidence, agreement and a hash of the inference result.

Accepted events are appended to a cell-specific dataset. Rejected events are quarantined as unknown. A cell-specific memory summarizes accepted musical events without silently retraining the identity model.

The evidence dashboard displays:

• current DSSC identity and authorization;
• Conv1D confidence;
• repeatability and physical-fingerprint agreement;
• the complete I-V curve;
• all sixteen resistor points;
• Voc, Isc, FF and maximum power;
• Rs*, Rsh*, curvature and area;
• the event identifier and hash;
• the World Radio Mixtable mapping;
• country, station, play and volume controls.

The dashboard is evidence for the local prototype. It is not evidence that a renewable-energy certificate was issued by an external authority.

World Radio Mixtable

Purpose

World Radio Mixtable converts an accepted I-V measurement into a six-channel internet-radio composition. It makes the electrical fingerprint audible while keeping the mapping inspectable.

The same quantized features create the same channel table and mix hash. Material drift can change the levels or move a country selection across a defined boundary, but the system does not make an unconstrained random playlist.

Manufacturing a mixtable

The mixtable is manufactured in five stages:

1. Extract the physical features from the accepted curve.
2. Quantize the feature values used by the audio mapping.
3. Normalize each selected feature inside a fixed range.
4. Convert normalized values into channel volume and country buckets.
5. Resolve one radio station for each country/genre pair and record a mix hash.

The resulting mixtable records the source feature, physical range, normalized value, volume, semantic interpretation, country-selection feature, selected country and station for every channel.

Six-channel mapping

Channel	Volume source	Musical interpretation	Country source
Techno	Isc	Raw current drive	Normalized area
Electronic	Curvature	Edge sharpness	Curvature
Jazz	Fill factor	Knee structure and swing	Fill factor
News	Voc	Announced potential	Voc
Rock	Rs*	Series grit	Rs*
Ambient	Rsh*	Shunt-leak atmosphere	Rsh*

Each volume is derived from a fixed feature range. Very low channels are muted. The master volume is also influenced by identity confidence.

Station discovery

The application can search the Radio Browser directory by genre and selected country. Resolved stations are cached so repeated measurements remain musically stable. Offline fallback streams allow the installation to demonstrate its identity logic without depending on a fresh search result.

Playback and use

After successful verification, the local player starts one background VLC process per channel. A localhost audio and control service exposes play, pause, status and volume operations to the browser dashboard.

The listener can:

• hear all six feature-derived stations together;
• pause or resume a channel;
• adjust a channel volume while retaining the measured default;
• inspect why that channel received its volume and country;
• compare the sonic balance of different DSSCs.

A new physical button press stops the previous six-channel session before creating the next one. Rejection produces silence instead of borrowing the identity or mix of another cell.

Manual World Radio Mixer application

A separate manual World Radio Mixer application also exists for exploring internet-radio combinations without DSSC identity. It can create up to eight manually selected decks. It should be distinguished from the six-channel Solar Oracle Mixtable, which is generated deterministically from a verified physical measurement.

Application inventory

Application or method	Role
I-V tester firmware	Performs the physical sixteen-load sweep, button workflow, local display and push request.
Conv1D Oracle tools	Collect measurements, train or verify one-cell models, evaluate the two-cell classifier and create authorization reports.
Solar Oracle receiver	Receives the physical press, runs identity, records the event, updates evidence and orchestrates sound.
World Radio Mixtable builder	Converts photovoltaic features into six deterministic channel mappings.
Mixtable player	Runs six VLC channels and the local browser control service.
Evidence dashboard generator	Produces the live 20 by 20 evidence mosaic.
E-paper status output	Produces a compact physical-display representation of identity and authorization.
Manual World Radio Mixer	Provides independent manual exploration of multi-station mixes.
Max / Gen performance patch	Receives optional OSC controls for local synthesis and performance experiments.
Local dataset and memory tools	Inspect identity-linked events and create deterministic replay-control records.
Smart-contract prototype	Optionally verifies signed inference reports; not active in the present installation.
RAVE dataset scaffold	Accumulates identity-linked source events for future neural-audio research.

Current MVP

The current minimum viable prototype demonstrates that:

1. a physical DSSC measurement can initiate a complete digital event;
2. two locally enrolled cells can be classified under controlled conditions;
3. out-of-domain measurements can be rejected as unknown;
4. a verified measurement can create a deterministic six-radio composition;
5. the electrical-to-musical mapping can remain visible and auditable;
6. rejection can produce silence;
7. local hash-linked evidence can connect material measurement, identity and sound.

This is described as minimal green-credential sonification: the system makes a local eligibility decision audible, but it does not issue a certified green credential.

RAVE roadmap

The current output is an internet-radio mixtable, not neural audio. The next research stage is to turn accepted, licensed audio selections into an identity-linked dataset and train a RAVE model in the cloud.

The intended transition is:

current MVP:
Conv1D + 7D evidence -> six-station World Radio Mixtable

future:
Conv1D identity + 8D learned latent control
 -> licensed identity-linked audio dataset
 -> trained RAVE model
 -> reproducible generative sound

The radio-search, licensing, dataset preparation and cloud-training services may later be operated by software agents. Agent-to-agent protocols are only expected to become useful when these services are autonomous and operated across separate platforms or organisations. They are not required for the present local MVP.

Long-term goal: distributed green credentials

The long-term aim is not simply to convert voltage into music. It is to investigate infrastructure in which a material energy source can:

• establish a repeatable physical identity;
• sign or authorize an event through measured behaviour;
• maintain a provenance chain without exposing unnecessary raw identity data;
• delegate limited actions to an energy or software agent;
• participate in a distributed green-credential system;
• make authorization and rejection publicly perceptible through sound.

Reaching this goal requires larger datasets, calibrated illumination, formal PUF evaluation, safe model updates, licensed audio data, active external registries and public transaction evidence.

Limitations

• The current identity study contains only two enrolled DSSCs.
• The main validation was performed in one tester domain under controlled but uncalibrated illumination.
• Contact placement, angle, distance, ageing and temperature can change the curve.
• Cross-instrument electrochemical measurements cannot be treated as direct inputs to the photovoltaic resistor-sweep model.
• The open-set thresholds are prototype thresholds, not population-level security limits.
• Local registration is not an active blockchain or renewable-certificate registration.
• The current dashboard's green-credential status is a local artistic rule.
• No trained RAVE model is active.
• Radio and future training data require explicit licensing and provenance review.

Credits

• Solar Oracle Walkman: ongoing project by Shih Wei Chieh and collaborators.
• I-V curve tester foundation: Marc Dusseiller.
• DSSC measurement, machine identity, radio sonification and green-credential research developed as an open, ongoing prototype.

Source

• Solar Oracle Walkman / I-V Curve Tester
• The original idea on September 2025: The Solar Oracle Walkman